Merge local master 5135a036 into remed/lane-AbLegacy

Refs #738
Claude-Session: https://claude.ai/code/session_01Ub9hGmYLQvn9XETzw7CMsT

docs/DriverLifecycle.md

@@ -512,16 +512,16 @@ same pass (a UNS reference variable, say) are visible through the index with no
 
 ⚠️ **The value fan is now a single message.** It sends one
 `DriverInstanceActor.AttributeValueBatchPublished` carrying one item per RawPath, where it previously sent
-one `AttributeValuePublished` per ref — 5 000 messages onto the mux that also carries the node's steady
+one per-value message per ref — 5 000 messages onto the mux that also carries the node's steady
 value ingress, every time one driver flapped. Delivery is unchanged: the mux filters the batch per item
 against the same registered-interest map the per-value receive consulted, so only the framing moved. The
 **alarm** fan is indexed but deliberately not batched — one driver ref can back several condition NodeIds,
 so it still emits one `AlarmQualityUpdate` per condition.
 
-**Consequence worth knowing: `AttributeValuePublished` now has no production sender at all.** This fan was
-its last one — everything on the subscription path was already wrapped in `AttributeValueBatchPublished`.
-`DependencyMuxActor` still registers the per-value receive, but in production it is dead code; it stays
-for tests and direct senders.
+**Consequence: the per-value `AttributeValuePublished` record is deleted.** This fan was its last
+production sender (everything on the subscription path was already wrapped in
+`AttributeValueBatchPublished`), so the record and `DependencyMuxActor`'s per-value receive were removed
+(2026-10-22). A single value is a one-item batch; tests build one with `ValuePublishes.Single`.
 
 #### The third fan: the value nodes (archreview 03/S15)
 

docs/ReadWriteOperations.md

@@ -39,8 +39,8 @@ This was not always true. Until #519 the publish path projected the driver's `ui
 `Good`/`Uncertain`/`Bad` enum in `DriverInstanceActor`, and `OtOpcUaNodeManager` re-expanded `Bad` to
 a bare `StatusCodes.Bad` (`0x80000000`). Every driver's diagnostic intent survived in logs only —
 including the 16 status constants #497 had just corrected across six drivers. The value path now
-carries the raw `uint` end to end (`AttributeValuePublished` → `AttributeValueUpdate` →
-`IOpcUaAddressSpaceSink.WriteValue`), and severity classification is a helper
+carries the raw `uint` end to end (`AttributeValueBatchPublished` → `AttributeValueBatchUpdate` →
+`IOpcUaAddressSpaceSink.WriteValues`), and severity classification is a helper
 (`OpcUaStatus.IsGood`/`IsUncertain`/`IsBad`) rather than a lossy projection.
 
 **Writing a status code in engine code:** use a named constant on

docs/ScriptEditor.md

@@ -198,10 +198,32 @@ scripted-alarm predicates) runs scripts through the shared `ScriptRunGuard`:
   2 s default) quarantines immediately, through the same path (Error log, metric
   `reason=stuck`, Bad). A short overrun (a young orphan) stays a plain refusal and
   the script resumes once the orphan completes.
+- **Cooperative cancellation (`ctx.CancellationToken`).** The guard cancels the run
+  token when a run times out, and scripts see it as `ctx.CancellationToken`. A
+  long-running loop that polls `ctx.CancellationToken.IsCancellationRequested` (or calls
+  `ctx.CancellationToken.ThrowIfCancellationRequested()`) ends its own abandoned run
+  and frees the slot, instead of holding a pool thread until it returns. A script that
+  ignores the token is still an orphan. The editor completes the member, and
+  `CancellationToken` is on the sandbox allow-list.
 - **Allocation budget.** A single run allocating more than
   `Scripting:MaxAllocBytesPerRun` (64 MB) is quarantined. This is sampled after the
   run returns, so a run that OOMs the process first is not stopped — that needs an
   out-of-process runner.
+- **Configuration keys and their ranges.** All three knobs are node-wide (the
+  `Scripting` appsettings section, shared by virtual tags, Calculation tags and
+  scripted alarms) and are clamped when bound, with one startup Warning per
+  clamped value:
+
+  | Key | Default | Range after the clamp |
+  |---|---|---|
+  | `Scripting:MaxAllocBytesPerRun` | `67108864` (64 MB) | `0` (or negative) disables the budget; a positive value below `1048576` (1 MB) becomes 1 MB |
+  | `Scripting:QuarantineAfterConsecutiveTimeouts` | `3` | `1`–`1000` |
+  | `Scripting:StuckOrphanQuarantineFactor` | `10` | `0` (or negative) disables the rule; above `1000` becomes `1000`; `NaN` becomes the default |
+
+  A Calculation driver may override `maxAllocBytesPerRun` and
+  `quarantineAfterConsecutiveTimeouts` in its own driver config. Those keys are
+  layered on the node values and clamped to the same ranges, and an out-of-range
+  value logs a Warning naming the driver. The stuck factor has no per-driver key.
 - **Lifting.** A quarantine lifts when the script source changes or on redeploy,
   never by itself.
 - **Script-log rate limit.** The cluster-wide script-log fan-out admits 20 entries
@@ -243,7 +265,7 @@ prefix. The concrete implementation `ScriptTagCatalog` reads `Tag` +
 | Virtual tag | Leaf `Name` (best-effort) |
 
 **Why only resolvable keys?** The live runtime resolves a `ctx.GetTag("X")`
-literal against `DriverInstanceActor.AttributeValuePublished.FullReference`,
+literal against `DriverInstanceActor.AttributeValueItem.FullReference`,
 which is the `FullName` field from `Tag.TagConfig`
 (see `ScriptTagCatalog.ExtractFullNameFromTagConfig` + `EquipmentNodeWalker.ExtractFullName`).
 The former UNS-path cascade engine (`Core.VirtualTags.VirtualTagEngine`, keyed by the

docs/VirtualTags.md

@@ -93,7 +93,7 @@ Virtual-tag nodes are materialised by `AddressSpaceApplier` and fed by `VirtualT
 
 `ITagUpstreamSource` and `IHistoryWriter` are the two ports value consumers require from their host. Both live in `Core.VirtualTags`. In the v2 actor system:
 
-- **Upstream-tag feed.** `DependencyMuxActor` (`src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs`) routes `DriverInstanceActor.AttributeValuePublished` events to the `VirtualTagActor` instances that declared interest in those tag refs. Each `VirtualTagActor` holds the in-memory per-tag dependency map; the `IVirtualTagEvaluator` (`RoslynVirtualTagEvaluator`) receives the dependency snapshot synchronously on the actor message thread. Reads of never-pushed dependency refs return `null` values in the dependency snapshot.
+- **Upstream-tag feed.** `DependencyMuxActor` (`src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs`) routes the items of each `DriverInstanceActor.AttributeValueBatchPublished` to the `VirtualTagActor` instances that declared interest in those tag refs. Each `VirtualTagActor` holds the in-memory per-tag dependency map; the `IVirtualTagEvaluator` (`RoslynVirtualTagEvaluator`) receives the dependency snapshot synchronously on the actor message thread. Reads of never-pushed dependency refs return `null` values in the dependency snapshot.
 - **`IHistoryWriter`** — the equipment-namespace path threads `Historize` end-to-end and `VirtualTagHostActor` invokes the injected writer on historized results (H5); the writer is resolved through `DriverHostActor` DI with a `NullHistoryWriter` default. No *durable* writer ships on this seam — durable continuous historization now flows through the separate `ContinuousHistorizationRecorder` → HistorianGateway `WriteLiveValues` path (see the `IHistoryWriter` section above and [Historian.md](Historian.md)).
 
 ## Composition
@@ -104,7 +104,7 @@ In the v2 actor system, virtual-tag engine composition is owned by the driver-ro
 
 - `AddressSpaceComposer.Compose` emits `DriverInstancePlan` / `ScriptedAlarmPlan` records; the driver-role `DriverHostActor` spawns one `VirtualTagActor` per virtual-tag expression and one `ScriptedAlarmActor` per scripted alarm.
 - `RoslynVirtualTagEvaluator` (`src/Server/ZB.MOM.WW.OtOpcUa.Host/Engines/RoslynVirtualTagEvaluator.cs`) is injected into each `VirtualTagActor` as its `IVirtualTagEvaluator`. It holds a per-source `CompiledScriptCache` keyed by script source and compiles on first use.
-- `DependencyMuxActor` (`src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs`) receives every `DriverInstanceActor.AttributeValuePublished` event and routes it to the `VirtualTagActor` instances that registered interest in that tag ref.
+- `DependencyMuxActor` (`src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs`) receives every `DriverInstanceActor.AttributeValueBatchPublished` and routes each item to the `VirtualTagActor` instances that registered interest in that tag ref.
 
 `ITagUpstreamSource` and `IHistoryWriter` remain available as `Core.VirtualTags` ports.
 
@@ -125,7 +125,7 @@ In the v2 actor system, virtual-tag engine composition is owned by the driver-ro
 - `src/Core/ZB.MOM.WW.OtOpcUa.Core.VirtualTags/ITagUpstreamSource.cs` — driver-tag read + subscribe port
 - `src/Core/ZB.MOM.WW.OtOpcUa.Core.VirtualTags/IHistoryWriter.cs` — historize sink port + `NullHistoryWriter`
 - `src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/VirtualTagActor.cs` — actor that receives `DependencyValueChanged` from the mux and invokes `IVirtualTagEvaluator` per expression
-- `src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs` — routes `DriverInstanceActor.AttributeValuePublished` to interested `VirtualTagActor` subscribers
+- `src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs` — routes `DriverInstanceActor.AttributeValueBatchPublished` items to interested `VirtualTagActor` subscribers
 - `src/Server/ZB.MOM.WW.OtOpcUa.Host/Engines/RoslynVirtualTagEvaluator.cs` — production `IVirtualTagEvaluator` binding; holds a per-source `CompiledScriptCache`
 - `src/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer/AddressSpaceComposer.cs` — pure data composer: config-DB entities → `AddressSpaceComposition` (UNS topology + raw subtree + driver/alarm plans)
 - `src/Server/ZB.MOM.WW.OtOpcUa.OpcUaServer/AddressSpaceApplier.cs` — applies the composed plan into the SDK node manager

src/Core/ZB.MOM.WW.OtOpcUa.Core.Abstractions/IDependencyConsumer.cs

@@ -11,7 +11,7 @@ namespace ZB.MOM.WW.OtOpcUa.Core.Abstractions;
 ///     <para>
 ///         The mechanism mirrors how a <c>VirtualTagActor</c> registers interest on the
 ///         <c>DependencyMuxActor</c>: the mux already receives every driver's
-///         <c>AttributeValuePublished</c> keyed by wire-ref (under v3, RawPath), so a driver that
+///         published values (batched, one item per value) keyed by wire-ref (under v3, RawPath), so a driver that
 ///         implements this interface simply piggy-backs on that fan-out. A calc tag's computed output
 ///         re-enters the mux via the ordinary driver publish path, so calc-of-calc chains work with no
 ///         extra machinery — which is exactly why the deploy-time cycle gate is mandatory.

src/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting.Abstractions/ScriptContext.cs

@@ -80,6 +80,50 @@ public abstract class ScriptContext
     /// </summary>
     public abstract ILogger Logger { get; }
 
+    /// <summary>
+    ///     The current run's cancellation token. The run guard cancels it when the run exceeds its wall-clock
+    ///     budget, and it is linked to the caller's token, so it is also cancelled when the caller cancels (host
+    ///     shutdown, an actor stopping mid-run). A long-running script can poll <c>ctx.CancellationToken.IsCancellationRequested</c> (or call
+    ///     <c>ThrowIfCancellationRequested()</c>) and end its own abandoned run instead of holding a pool thread
+    ///     until it returns. Cooperative only: a script that never looks at it still runs to completion as an
+    ///     orphan. <see cref="CancellationToken.None"/> outside a run.
+    /// </summary>
+    /// <remarks>
+    ///     Scoped per run through an <see cref="AsyncLocal{T}"/>, not stored on the context, because a context
+    ///     object can be reused across runs (the scripted-alarm engine keeps one per alarm). An abandoned orphan
+    ///     therefore keeps seeing its own cancelled token even after a newer run starts on the same context.
+    /// </remarks>
+    public CancellationToken CancellationToken => s_runToken.Value;
+
+    private static readonly AsyncLocal<CancellationToken> s_runToken = new();
+
+    /// <summary>
+    ///     Makes <paramref name="token"/> the script-visible <see cref="CancellationToken"/> on the calling
+    ///     execution context until the returned scope is disposed. Called by <c>ScriptEvaluator.RunAsync</c>
+    ///     around the compiled delegate.
+    /// </summary>
+    /// <param name="token">The run token.</param>
+    /// <returns>A scope that restores the previous token.</returns>
+    internal static RunTokenScope EnterRun(CancellationToken token)
+    {
+        var previous = s_runToken.Value;
+        s_runToken.Value = token;
+        return new RunTokenScope(previous);
+    }
+
+    /// <summary>Restores the token that was current before <see cref="EnterRun"/>.</summary>
+    internal readonly struct RunTokenScope : IDisposable
+    {
+        private readonly CancellationToken _previous;
+
+        /// <summary>Captures the token to restore.</summary>
+        /// <param name="previous">The token current before the run.</param>
+        public RunTokenScope(CancellationToken previous) => _previous = previous;
+
+        /// <inheritdoc />
+        public void Dispose() => s_runToken.Value = _previous;
+    }
+
     /// <summary>
     ///     Deadband helper — returns <c>true</c> when <paramref name="current"/> differs
     ///     from <paramref name="previous"/> by more than <paramref name="tolerance"/>.

src/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting.Abstractions/ZB.MOM.WW.OtOpcUa.Core.Scripting.Abstractions.csproj

@@ -16,4 +16,10 @@
     <ProjectReference Include="..\ZB.MOM.WW.OtOpcUa.Core.Abstractions\ZB.MOM.WW.OtOpcUa.Core.Abstractions.csproj"/>
   </ItemGroup>
 
+  <ItemGroup>
+    <!-- ScriptEvaluator scopes the script-visible run token (ScriptContext.EnterRun). Internal so a script
+         cannot replace its own token. -->
+    <InternalsVisibleTo Include="ZB.MOM.WW.OtOpcUa.Core.Scripting"/>
+  </ItemGroup>
+
 </Project>

src/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting/ScriptEvaluator.cs

@@ -185,7 +185,13 @@ public sealed class ScriptEvaluator<TContext, TResult> : IDisposable
         // TResult directly). We surface a Task<TResult> only to keep the existing
         // RunAsync contract consumers depend on. TimedScriptEvaluator wraps this in
         // Task.Run so a long-running script still honours its wall-clock budget.
-        var result = _func(globals);
+        // The run token is the script-visible ctx.CancellationToken for exactly this call. ScriptRunGuard passes
+        // the token it cancels on timeout, so a polling script can end its own abandoned run.
+        TResult result;
+        using (ScriptContext.EnterRun(ct))
+        {
+            result = _func(globals);
+        }
         return Task.FromResult(result);
     }
 

src/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting/ScriptRunGuard.cs

@@ -96,9 +96,11 @@ public sealed class ScriptRunGuard
     /// <param name="logger">Where the quarantine Error line goes; null ⇒ Serilog's global logger.</param>
     public ScriptRunGuard(ScriptingOptions? options = null, ILogger? logger = null)
     {
-        options ??= new ScriptingOptions();
+        // Clamped here as well as at the bind site, because options also arrive from a Calculation driver's own
+        // config and from tests. An unclamped factor made the refusal path's timeout * factor overflow.
+        options = (options ?? new ScriptingOptions()).Sanitized(out _);
         _maxAllocBytesPerRun = options.MaxAllocBytesPerRun;
-        _quarantineThreshold = Math.Max(1, options.QuarantineAfterConsecutiveTimeouts);
+        _quarantineThreshold = options.QuarantineAfterConsecutiveTimeouts;
         _stuckFactor = options.StuckOrphanQuarantineFactor;
         _logger = logger ?? Log.Logger;
     }
@@ -197,7 +199,11 @@ public sealed class ScriptRunGuard
             if (_stuckFactor > 0 && Volatile.Read(ref state.RunStartedTimestamp) is var startedAt and not 0)
             {
                 var running = Stopwatch.GetElapsedTime(startedAt);
-                var threshold = timeout * _stuckFactor;
+                // Saturating: even the clamped maximum factor times a huge timeout must not overflow TimeSpan.
+                var thresholdTicks = timeout.Ticks * _stuckFactor;
+                var threshold = thresholdTicks >= TimeSpan.MaxValue.Ticks
+                    ? TimeSpan.MaxValue
+                    : TimeSpan.FromTicks((long)thresholdTicks);
                 if (running > threshold)
                 {
                     Quarantine(key, state,

src/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting/ScriptingOptions.cs

@@ -53,4 +53,85 @@ public sealed class ScriptingOptions
     ///     refusal. Zero or negative disables the rule. Default 10.
     /// </summary>
     public double StuckOrphanQuarantineFactor { get; set; } = DefaultStuckOrphanQuarantineFactor;
+
+    /// <summary>Upper bound <see cref="Sanitized"/> applies to <see cref="StuckOrphanQuarantineFactor"/>.</summary>
+    public const double MaxStuckOrphanQuarantineFactor = 1000;
+
+    /// <summary>Upper bound <see cref="Sanitized"/> applies to <see cref="QuarantineAfterConsecutiveTimeouts"/>.</summary>
+    public const int MaxQuarantineAfterConsecutiveTimeouts = 1000;
+
+    /// <summary>
+    ///     Lower bound <see cref="Sanitized"/> applies to a positive <see cref="MaxAllocBytesPerRun"/>: 1 MB. A
+    ///     smaller budget would quarantine practically every script on its first run.
+    /// </summary>
+    public const long MinMaxAllocBytesPerRun = 1024L * 1024;
+
+    /// <summary>
+    ///     Returns a copy with every numeric knob pulled into a sane range. The bind site calls it and logs
+    ///     <paramref name="adjustments"/>. <see cref="ScriptRunGuard"/> calls it again, so options that reach the
+    ///     guard by another route (a Calculation driver's own config, a test) are clamped too.
+    ///     <list type="bullet">
+    ///         <item><see cref="StuckOrphanQuarantineFactor"/>: NaN becomes the default, a negative value becomes 0
+    ///         (disabled), and a value above <see cref="MaxStuckOrphanQuarantineFactor"/> becomes that maximum.
+    ///         Unclamped, an absurd factor made <c>timeout * factor</c> throw <see cref="OverflowException"/> on
+    ///         the first refused run.</item>
+    ///         <item><see cref="QuarantineAfterConsecutiveTimeouts"/>: into [1, <see cref="MaxQuarantineAfterConsecutiveTimeouts"/>].</item>
+    ///         <item><see cref="MaxAllocBytesPerRun"/>: a negative value becomes 0 (disabled), and a positive value
+    ///         below <see cref="MinMaxAllocBytesPerRun"/> becomes that minimum.</item>
+    ///     </list>
+    /// </summary>
+    /// <param name="adjustments">One line per value that was out of range, naming its key. A negative value that
+    /// already meant "disabled" is normalised without a line.</param>
+    /// <returns>A new, clamped instance. This instance is not modified.</returns>
+    public ScriptingOptions Sanitized(out IReadOnlyList<string> adjustments)
+    {
+        var notes = new List<string>();
+
+        var factor = StuckOrphanQuarantineFactor;
+        if (double.IsNaN(factor))
+        {
+            notes.Add($"{SectionName}:{nameof(StuckOrphanQuarantineFactor)} is NaN; using the default {DefaultStuckOrphanQuarantineFactor}.");
+            factor = DefaultStuckOrphanQuarantineFactor;
+        }
+        else if (factor < 0)
+        {
+            factor = 0;
+        }
+        else if (factor > MaxStuckOrphanQuarantineFactor)
+        {
+            notes.Add($"{SectionName}:{nameof(StuckOrphanQuarantineFactor)} {factor} is above the maximum; clamped to {MaxStuckOrphanQuarantineFactor}.");
+            factor = MaxStuckOrphanQuarantineFactor;
+        }
+
+        var threshold = QuarantineAfterConsecutiveTimeouts;
+        if (threshold < 1)
+        {
+            notes.Add($"{SectionName}:{nameof(QuarantineAfterConsecutiveTimeouts)} {threshold} is below 1; clamped to 1.");
+            threshold = 1;
+        }
+        else if (threshold > MaxQuarantineAfterConsecutiveTimeouts)
+        {
+            notes.Add($"{SectionName}:{nameof(QuarantineAfterConsecutiveTimeouts)} {threshold} is above the maximum; clamped to {MaxQuarantineAfterConsecutiveTimeouts}.");
+            threshold = MaxQuarantineAfterConsecutiveTimeouts;
+        }
+
+        var alloc = MaxAllocBytesPerRun;
+        if (alloc < 0)
+        {
+            alloc = 0;
+        }
+        else if (alloc > 0 && alloc < MinMaxAllocBytesPerRun)
+        {
+            notes.Add($"{SectionName}:{nameof(MaxAllocBytesPerRun)} {alloc} is below the minimum; clamped to {MinMaxAllocBytesPerRun}.");
+            alloc = MinMaxAllocBytesPerRun;
+        }
+
+        adjustments = notes;
+        return new ScriptingOptions
+        {
+            MaxAllocBytesPerRun = alloc,
+            QuarantineAfterConsecutiveTimeouts = threshold,
+            StuckOrphanQuarantineFactor = factor,
+        };
+    }
 }

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Contracts/AbCipDriverOptions.cs

@@ -117,6 +117,12 @@ public sealed class AbCipDriverOptions
     ///     device round trips (the whole point: a strictly sequential batch pays one full round
     ///     trip per tag); lowering it to <c>1</c> restores strictly sequential reads. Values below
     ///     <c>1</c> are clamped to <c>1</c> at the use site.
+    ///     <para>
+    ///     On the <see cref="AbCipTransport.Native"/> transport the effective window is at least 64: there the
+    ///     window caps how many services one Multiple Service Packet coalesces (the device's shared scheduler packs
+    ///     what is queued, bounded by the connection's request/reply byte budget), so a small value only multiplies
+    ///     round trips. A value above 64 is honoured; a value below it is raised.
+    ///     </para>
     /// </summary>
     public int MaxConcurrentReads { get; init; } = 8;
 

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipArrayCodec.cs

@@ -0,0 +1,80 @@
+using System.Buffers.Binary;
+using System.Collections;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+/// <summary>
+///     Encodes a whole-array write (review 07 U10): <c>count</c> elements at the element type's wire width,
+///     little-endian, exactly the layout a Logix atomic array holds and the read path decodes. Shared by both
+///     runtimes so the bytes on the wire are transport-independent.
+/// </summary>
+/// <remarks>
+///     <para><b>Element values CONVERT, they are not exact-matched.</b> ScadaBridge (and any OPC UA client that
+///     round-trips what it read) writes CLR lists such as <c>int[]</c> to an INT array — the read path itself
+///     publishes INT/SINT elements as <c>int</c> — so each element goes through the same <see cref="Convert"/> the
+///     scalar path uses. A value that does not fit is an <see cref="OverflowException"/> (BadOutOfRange) and one that
+///     cannot convert a <see cref="FormatException"/> / <see cref="InvalidCastException"/> (BadTypeMismatch), all
+///     thrown before any byte is sent.</para>
+///     <para>BOOL arrays (bit-packed DWORDs — a byte-per-element encode would clobber neighbour bits), STRING arrays
+///     (structures) and structure arrays are not encodable here; the driver refuses them <c>BadNotSupported</c>
+///     before a runtime exists.</para>
+/// </remarks>
+internal static class AbCipArrayCodec
+{
+    /// <summary>Wire width of one element of an encodable array type, or <c>null</c> when the type is not encodable.</summary>
+    /// <param name="type">The element type.</param>
+    /// <returns>The element width in bytes, or <c>null</c>.</returns>
+    public static int? ElementWidth(AbCipDataType type) => type switch
+    {
+        AbCipDataType.SInt or AbCipDataType.USInt => 1,
+        AbCipDataType.Int or AbCipDataType.UInt => 2,
+        AbCipDataType.DInt or AbCipDataType.UDInt or AbCipDataType.Real or AbCipDataType.Dt => 4,
+        AbCipDataType.LInt or AbCipDataType.ULInt or AbCipDataType.LReal => 8,
+        _ => null,
+    };
+
+    /// <summary>
+    ///     The elements of an array write value, or <c>null</c> when the value is not a list (a scalar, a string, null).
+    ///     Any non-string <see cref="IEnumerable"/> is accepted — <c>int[]</c>, <c>short[]</c>, <c>List&lt;object&gt;</c> …
+    /// </summary>
+    /// <param name="value">The client's write value.</param>
+    /// <returns>The elements, or <c>null</c>.</returns>
+    public static object?[]? Elements(object? value) => value switch
+    {
+        null or string => null,
+        IEnumerable list => [.. list.Cast<object?>()],
+        _ => null,
+    };
+
+    /// <summary>Encodes <paramref name="values"/> as <paramref name="type"/> elements.</summary>
+    /// <param name="type">The element type (must have an <see cref="ElementWidth"/>).</param>
+    /// <param name="values">The element values; their count is the element count.</param>
+    /// <returns>The little-endian wire bytes, <c>values.Count × ElementWidth(type)</c> long.</returns>
+    /// <exception cref="NotSupportedException">The type is not encodable.</exception>
+    public static byte[] Encode(AbCipDataType type, IReadOnlyList<object?> values)
+    {
+        var width = ElementWidth(type)
+            ?? throw new NotSupportedException($"Array writes of {type} elements are not supported.");
+        var buffer = new byte[values.Count * width];
+        for (var i = 0; i < values.Count; i++)
+        {
+            var v = values[i];
+            var span = buffer.AsSpan(i * width, width);
+            switch (type)
+            {
+                case AbCipDataType.SInt: span[0] = unchecked((byte)Convert.ToSByte(v)); break;
+                case AbCipDataType.USInt: span[0] = Convert.ToByte(v); break;
+                case AbCipDataType.Int: BinaryPrimitives.WriteInt16LittleEndian(span, Convert.ToInt16(v)); break;
+                case AbCipDataType.UInt: BinaryPrimitives.WriteUInt16LittleEndian(span, Convert.ToUInt16(v)); break;
+                case AbCipDataType.DInt:
+                case AbCipDataType.Dt: BinaryPrimitives.WriteInt32LittleEndian(span, Convert.ToInt32(v)); break;
+                case AbCipDataType.UDInt: BinaryPrimitives.WriteUInt32LittleEndian(span, Convert.ToUInt32(v)); break;
+                case AbCipDataType.LInt: BinaryPrimitives.WriteInt64LittleEndian(span, Convert.ToInt64(v)); break;
+                case AbCipDataType.ULInt: BinaryPrimitives.WriteUInt64LittleEndian(span, Convert.ToUInt64(v)); break;
+                case AbCipDataType.Real: BinaryPrimitives.WriteSingleLittleEndian(span, Convert.ToSingle(v)); break;
+                case AbCipDataType.LReal: BinaryPrimitives.WriteDoubleLittleEndian(span, Convert.ToDouble(v)); break;
+            }
+        }
+        return buffer;
+    }
+}

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipDriverProbe.cs

@@ -18,10 +18,11 @@ namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
 ///     non-CIP server) returns <c>Ok = false</c> with a "handshake failed" message instead of
 ///     a false-positive green tick.
 ///     <para>
-///     Status-code discrimination: <c>AbStatus.Ok</c> and <c>AbStatus.NotFound</c> mean the controller
-///     answered the CIP session (a non-existent probe tag name still proves CIP connectivity) and are
-///     treated as "reachable - Ok = true". Every other status (timeout, bad connection, bad reply,
-///     partial, abort, ...) is a transport or session failure: "handshake failed - Ok = false".
+///     Status-code discrimination: <c>AbStatus.Ok</c> and every device refusal (not found, out of bounds,
+///     not allowed, too large/small, type mismatch, unsupported) mean the controller answered the CIP session
+///     (a non-existent probe tag name still proves CIP connectivity) and are treated as "reachable - Ok = true".
+///     Every other status (timeout, bad connection, bad reply, partial, abort, ...) is a transport or session
+///     failure: "handshake failed - Ok = false".
 ///     </para>
 /// </summary>
 public sealed class AbCipDriverProbe : IDriverProbe
@@ -147,7 +148,7 @@ public sealed class AbCipDriverProbe : IDriverProbe
             {
                 var msg = statusCode == AbStatus.Ok
                     ? "CIP session OK"
-                    : "CIP session OK (controller reachable; probe tag not found)";
+                    : RefusedProbeMessage(statusCode);
                 return new(true, msg, sw.Elapsed);
             }
 
@@ -164,7 +165,7 @@ public sealed class AbCipDriverProbe : IDriverProbe
             if (IsReachableException(ex))
             {
                 sw.Stop();
-                return new(true, "CIP session OK (controller reachable; probe tag not found)", sw.Elapsed);
+                return new(true, RefusedProbeMessage(ex.Status), sw.Elapsed);
             }
 
             return new(false, $"Reachable at {host}:{port} but CIP handshake failed: {ex.Message}", null);
@@ -180,13 +181,20 @@ public sealed class AbCipDriverProbe : IDriverProbe
     }
 
     /// <summary>
-    ///     Returns <c>true</c> for transport statuses that indicate the CIP controller answered
-    ///     but the probe tag name was not found or did not match — sufficient to confirm CIP
-    ///     reachability. Returns <c>true</c> for <c>Ok</c> (tag was found).
+    ///     Returns <c>true</c> for transport statuses that indicate the CIP controller answered — <c>Ok</c>, or any
+    ///     device refusal (<see cref="AbCipRefusals.IsDeviceRefusal(AbStatus)"/>: not found, out of bounds, not
+    ///     allowed, too large/small, type mismatch, unsupported) — sufficient to confirm CIP reachability.
     /// </summary>
     /// <param name="s">The transport status to classify.</param>
     /// <returns><see langword="true"/> if the status confirms CIP reachability.</returns>
-    internal static bool IsReachableStatus(AbStatus s) => s is AbStatus.Ok or AbStatus.NotFound;
+    internal static bool IsReachableStatus(AbStatus s) => s == AbStatus.Ok || AbCipRefusals.IsDeviceRefusal(s);
+
+    /// <summary>The Test-Connect message for a controller that answered but refused the probe tag.</summary>
+    /// <param name="status">The refusal.</param>
+    /// <returns>The message.</returns>
+    private static string RefusedProbeMessage(AbStatus status) => status == AbStatus.NotFound
+        ? "CIP session OK (controller reachable; probe tag not found)"
+        : $"CIP session OK (controller reachable; probe tag refused: {status})";
 
     /// <summary>
     ///     Decides whether an <see cref="AbTransportException"/> is a "controller answered CIP but the
@@ -198,7 +206,8 @@ public sealed class AbCipDriverProbe : IDriverProbe
     internal static bool IsReachableException(AbTransportException ex)
     {
         ArgumentNullException.ThrowIfNull(ex);
-        return IsReachableStatus(ex.Status);
+        // A local refusal (the library / driver refused without any I/O) proves nothing about the controller.
+        return ex is not AbCipLocalRefusalException && IsReachableStatus(ex.Status);
     }
 
     private static (AbCipHostAddress? parsed, AbCipDeviceOptions? device) ExtractTarget(AbCipDriverOptions opts)

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipInitializationSupersededException.cs

@@ -0,0 +1,18 @@
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+/// <summary>
+///     Thrown by <see cref="AbCipDriver.InitializeAsync"/> when a shutdown, dispose or newer initialization retired the
+///     driver's tables while this initialization was building, so it published nothing. Deliberately NOT an
+///     <see cref="OperationCanceledException"/>: nobody cancelled this call, and a caller that treats cancellation as
+///     "my own token fired" must not mistake a superseded attempt for one. The superseding operation owns the driver's
+///     state; the caller (normally the runtime, which has already abandoned this attempt) can ignore the result.
+/// </summary>
+/// <remarks>
+///     Resilience accounting (#737 m2): not being an OCE, it passes the Initialize pipeline's breaker
+///     <c>ShouldHandle</c> (<c>DriverResiliencePipelineBuilder</c>) and, if its generation still matched, the actor's
+///     <c>RecordFault</c>. That is practically unreachable — <c>DriverInstanceActor</c> runs initialisation single-flight
+///     and drops generation-mismatched results — and a superseded attempt is at worst one counted failure. Recognising
+///     it there is a runtime-side (Lane A) change if it is ever wanted.
+/// </remarks>
+/// <param name="message">The detail.</param>
+public sealed class AbCipInitializationSupersededException(string message) : InvalidOperationException(message);

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipLocalRefusalException.cs

@@ -0,0 +1,15 @@
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+/// <summary>
+///     A refusal raised LOCALLY — by the driver's own runtime or by libplctag (<c>ErrorNotImplemented</c>) — before or
+///     without any reply from the controller. It carries <see cref="AbStatus.Unsupported"/> (so it maps to
+///     <c>BadNotSupported</c>) but, unlike a device refusal, proves nothing about the device: it keeps the runtime and
+///     writes no health, does NOT advance <c>LastSuccessfulRead</c>, records nothing in the connect backoff, and is not
+///     "reachable" for the probe or Test Connect (review 07 / Gitea #737 I2).
+/// </summary>
+/// <param name="message">The refusal detail.</param>
+/// <param name="inner">The underlying exception, if any.</param>
+public sealed class AbCipLocalRefusalException(string message, Exception? inner = null)
+    : AbTransportException(AbStatus.Unsupported, message, inner);

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipRefusals.cs

@@ -0,0 +1,42 @@
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
+
+/// <summary>
+///     The ONE classification of a transport status into "the controller answered" vs "the wire failed", shared by
+///     the data path, the probe loop, the Test-Connect probe and the native runtime's initialisation.
+/// </summary>
+/// <remarks>
+///     A device refusal is deterministic: the session works and the controller said no to this one tag (a renamed
+///     tag, <c>Arr[999]</c>, a read-protected tag, a type the authored config does not match). It is per-tag Bad,
+///     keeps the tag runtime (evicting re-creates it — and, for a sole lease holder, re-Forward-Opens — on every poll),
+///     never writes <c>Degraded</c>, never opens the device-wide connect backoff, and warns once per tag. Everything
+///     else (timeout, bad connection, a reply that failed to parse, partial, abort, …) is a comms fault.
+///     <para>
+///     The native transport currently folds CIP general statuses 0x1C, 0x20 and 0xFF/0x2104 into
+///     <see cref="AbStatus.BadReply"/> (<c>AbStatusCodes.FromCip</c>, owned by <c>Driver.Ab.Transport</c>), which
+///     this classification cannot tell from a garbled reply; until the transport maps them explicitly they stay
+///     comms faults.
+///     </para>
+/// </remarks>
+internal static class AbCipRefusals
+{
+    /// <summary>True when <paramref name="status"/> proves the controller answered but refused this tag.</summary>
+    /// <param name="status">The transport status.</param>
+    /// <returns><see langword="true"/> for a deterministic device refusal.</returns>
+    public static bool IsDeviceRefusal(AbStatus status) => status is AbStatus.NotFound or AbStatus.OutOfBounds
+        or AbStatus.NotAllowed or AbStatus.TooLarge or AbStatus.TooSmall or AbStatus.TypeMismatch
+        or AbStatus.Unsupported;
+
+    /// <summary>True when <paramref name="status"/> (an <see cref="AbStatus"/> value) is a device refusal.</summary>
+    /// <param name="status">The runtime status word.</param>
+    /// <returns><see langword="true"/> for a deterministic device refusal.</returns>
+    public static bool IsDeviceRefusal(int status) => IsDeviceRefusal((AbStatus)status);
+
+    /// <summary>True when <paramref name="ex"/> is an <see cref="AbTransportException"/> carrying a device refusal.</summary>
+    /// <param name="ex">The exception.</param>
+    /// <returns><see langword="true"/> for a deterministic device refusal.</returns>
+    /// <remarks>An <see cref="AbCipLocalRefusalException"/> is never a device refusal: nothing reached the device.</remarks>
+    public static bool IsDeviceRefusal(Exception ex) =>
+        ex is AbTransportException ate and not AbCipLocalRefusalException && IsDeviceRefusal(ate.Status);
+}

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipStatusMapper.cs

@@ -15,6 +15,7 @@ public static class AbCipStatusMapper
     public const uint GoodMoreData = 0x00A60000u;
     public const uint BadInternalError = 0x80020000u;
     public const uint BadNodeIdUnknown = 0x80340000u;
+    public const uint BadNotReadable = 0x803A0000u;
     public const uint BadNotWritable = 0x803B0000u;
     public const uint BadOutOfRange = 0x803C0000u;
     public const uint BadNotSupported = 0x803D0000u;
@@ -51,4 +52,14 @@ public static class AbCipStatusMapper
     /// <returns>The OPC UA status code.</returns>
     public static uint MapException(Exception ex) =>
         AbOpcUaStatusMap.MapException(ex);
+
+    /// <summary>
+    ///     Status for a READ the controller refused: as <see cref="MapStatus(AbStatus)"/>, except that a
+    ///     <see cref="AbStatus.NotAllowed"/> answer to a read is <see cref="BadNotReadable"/> (the shared table's
+    ///     <c>BadNotWritable</c> describes a refused write).
+    /// </summary>
+    /// <param name="status">The refusal status.</param>
+    /// <returns>The OPC UA status code.</returns>
+    public static uint MapReadStatus(AbStatus status) =>
+        status == AbStatus.NotAllowed ? BadNotReadable : MapStatus(status);
 }

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipUdtMemberLayout.cs

@@ -25,6 +25,30 @@ namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip;
 /// </remarks>
 public static class AbCipUdtMemberLayout
 {
+    /// <summary>Layouts already computed, keyed by the parent definition's member list (reference identity).</summary>
+    private static readonly System.Runtime.CompilerServices.ConditionalWeakTable<IReadOnlyList<AbCipStructureMember>, LayoutBox> Cache = new();
+
+    /// <summary>Holds a computed layout, including a "not groupable" (<c>null</c>) verdict.</summary>
+    private sealed class LayoutBox(IReadOnlyDictionary<string, int>? offsets)
+    {
+        public IReadOnlyDictionary<string, int>? Offsets { get; } = offsets;
+    }
+
+    /// <summary>
+    ///     <see cref="TryBuild"/>, computed once per member list (review 07 P11: the read planner rebuilt the layout —
+    ///     a dictionary allocation and an O(members) walk — for every member reference on every poll, O(members²) per
+    ///     parent per tick). A parent definition's member list is immutable for the life of the tag table that holds
+    ///     it, so caching by list identity is exact; a re-init builds new lists and the old entries are collected with
+    ///     them (<see cref="System.Runtime.CompilerServices.ConditionalWeakTable{TKey,TValue}"/>).
+    /// </summary>
+    /// <param name="members">The parent definition's declared members.</param>
+    /// <returns>The member offsets, or <c>null</c> when the members cannot be laid out by declaration.</returns>
+    public static IReadOnlyDictionary<string, int>? GetOrBuild(IReadOnlyList<AbCipStructureMember> members)
+    {
+        ArgumentNullException.ThrowIfNull(members);
+        return Cache.GetValue(members, m => new LayoutBox(TryBuild(m))).Offsets;
+    }
+
     /// <summary>
     ///     Try to compute member offsets for the supplied declared members. Returns <c>null</c>
     ///     if any member type is unsupported for declaration-only layout.

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/AbCipUdtReadPlanner.cs

@@ -67,7 +67,7 @@ public static class AbCipUdtReadPlanner
                 continue;
             }
 
-            var offsets = AbCipUdtMemberLayout.TryBuild(parent.Members);
+            var offsets = AbCipUdtMemberLayout.GetOrBuild(parent.Members);
             if (offsets is null || !offsets.TryGetValue(memberName, out var offset))
             {
                 fallback.Add(new AbCipUdtReadFallback(i, name));

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/IAbCipTagRuntime.cs

@@ -85,6 +85,17 @@ public interface IAbCipTagRuntime : IDisposable
     /// <param name="bitIndex">Bit index for BOOL-within-DINT insertion, or null.</param>
     /// <param name="value">Value to encode.</param>
     void EncodeValue(AbCipDataType type, int? bitIndex, object? value);
+
+    /// <summary>
+    ///     Encode a WHOLE-array write of <paramref name="values"/> (<c>values.Count</c> elements of
+    ///     <paramref name="type"/>, at the element width — see <see cref="AbCipArrayCodec"/>) into the local buffer.
+    ///     Callers pair this with <see cref="WriteAsync"/>. A runtime that cannot write arrays throws
+    ///     <see cref="NotSupportedException"/>.
+    /// </summary>
+    /// <param name="type">The array element type.</param>
+    /// <param name="values">The element values; already length-checked against the tag's element count.</param>
+    void EncodeArray(AbCipDataType type, IReadOnlyList<object?> values) =>
+        throw new NotSupportedException($"{GetType().Name} does not support array writes.");
 }
 
 /// <summary>

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/LibplctagStatusTranslation.cs

@@ -14,11 +14,21 @@ internal static class LibplctagStatusTranslation
         Status.Ok => AbStatus.Ok,
         Status.Pending => AbStatus.Pending,
         Status.ErrorTimeout => AbStatus.Timeout,
-        Status.ErrorNotFound or Status.ErrorNoMatch or Status.ErrorBadDevice => AbStatus.NotFound,
+        // #737 I2: only ErrorNotFound is the controller saying "no such tag". ErrorNoMatch / ErrorBadDevice describe a
+        // session or device-level failure (no proof the controller answered this tag), so they are comms-class: with
+        // the widened refusal classification, folding them into NotFound made a broken session read as "reachable".
+        Status.ErrorNotFound => AbStatus.NotFound,
+        Status.ErrorNoMatch or Status.ErrorBadDevice => AbStatus.BadConnection,
         Status.ErrorNotAllowed => AbStatus.NotAllowed,
         Status.ErrorOutOfBounds => AbStatus.OutOfBounds,
         Status.ErrorTooLarge => AbStatus.TooLarge,
         Status.ErrorTooSmall => AbStatus.TooSmall,
+        // ErrorNotImplemented is the LIBRARY refusing (no I/O); the exception paths raise it as an
+        // AbCipLocalRefusalException so it never counts as "the device answered" (see FromException).
+        // ASSUMPTION (#737 m3): libplctag reports a failed operation by THROWING, so a NotImplemented reaches the driver
+        // only through FromException. AbStatus has no "local" member (it is Driver.Ab.Transport's, Lane K), so a
+        // NotImplemented observed only as a STATUS WORD (GetStatus after a non-throwing op) would still classify as a
+        // device Unsupported refusal; that combination is not produced by libplctag.NET 1.5.2.
         Status.ErrorUnsupported or Status.ErrorNotImplemented => AbStatus.Unsupported,
         Status.ErrorPartial => AbStatus.Partial,
         Status.ErrorAbort => AbStatus.Abort,
@@ -70,9 +80,21 @@ internal static class LibplctagStatusTranslation
     public static async Task Wrap(Func<Task> op)
     {
         try { await op().ConfigureAwait(false); }
-        catch (LibPlcTagException ex) { throw new AbTransportException(ToAbStatus(ex), ex.Message, ex); }
+        catch (LibPlcTagException ex) { throw FromException(ex); }
     }
 
+    /// <summary>
+    ///     The transport exception for a libplctag failure: <c>ErrorNotImplemented</c> becomes an
+    ///     <see cref="AbCipLocalRefusalException"/> (a library refusal, no I/O); everything else an
+    ///     <see cref="AbTransportException"/> carrying the translated status.
+    /// </summary>
+    /// <param name="ex">The libplctag exception.</param>
+    /// <returns>The exception to throw.</returns>
+    public static AbTransportException FromException(LibPlcTagException ex) =>
+        Enum.TryParse<Status>((ex.Message ?? string.Empty).Trim(), ignoreCase: true, out var s) && s == Status.ErrorNotImplemented
+            ? new AbCipLocalRefusalException($"libplctag does not implement this operation: {ex.Message}", ex)
+            : new AbTransportException(ToAbStatus(ex), ex.Message ?? string.Empty, ex);
+
     /// <summary>Runs <paramref name="op"/>, rethrowing any <see cref="LibPlcTagException"/> as <see cref="AbTransportException"/>.</summary>
     /// <typeparam name="T">The result type.</typeparam>
     /// <param name="op">The libplctag operation.</param>
@@ -80,7 +102,7 @@ internal static class LibplctagStatusTranslation
     public static async Task<T> Wrap<T>(Func<Task<T>> op)
     {
         try { return await op().ConfigureAwait(false); }
-        catch (LibPlcTagException ex) { throw new AbTransportException(ToAbStatus(ex), ex.Message, ex); }
+        catch (LibPlcTagException ex) { throw FromException(ex); }
     }
 
     /// <summary>Runs a synchronous libplctag call, rethrowing <see cref="LibPlcTagException"/> as <see cref="AbTransportException"/>.</summary>
@@ -90,7 +112,7 @@ internal static class LibplctagStatusTranslation
     public static T WrapSync<T>(Func<T> op)
     {
         try { return op(); }
-        catch (LibPlcTagException ex) { throw new AbTransportException(ToAbStatus(ex), ex.Message, ex); }
+        catch (LibPlcTagException ex) { throw FromException(ex); }
     }
 
     /// <summary>Runs a synchronous libplctag call, rethrowing <see cref="LibPlcTagException"/> as <see cref="AbTransportException"/>.</summary>
@@ -98,6 +120,6 @@ internal static class LibplctagStatusTranslation
     public static void WrapSync(Action op)
     {
         try { op(); }
-        catch (LibPlcTagException ex) { throw new AbTransportException(ToAbStatus(ex), ex.Message, ex); }
+        catch (LibPlcTagException ex) { throw FromException(ex); }
     }
 }

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/LibplctagTagRuntime.cs

@@ -29,8 +29,19 @@ internal sealed class LibplctagTagRuntime : IAbCipTagRuntime
     public LibplctagTagRuntime(AbCipTagCreateParams p)
     {
         _tag = BuildTag(p);
+        _boolArrayElement = !p.IsArray && p.AuthoredType == AbCipDataType.Bool
+            && NativeAbCipTagRuntime.TrySplitBoolArrayElement(p.TagName, out _, out _);
     }
 
+    /// <summary>
+    ///     A scalar BOOL authored over a BOOL-array element (<c>Bits[n]</c>). libplctag reads it as whatever the
+    ///     controller returns for that element segment — a packed DWORD — and exposes no type code, so which DWORD
+    ///     came back (and so which bit is element n) cannot be checked here. Review 07 U11: such a read used to decode
+    ///     "low byte != 0" as Good; it is refused (<c>BadNotSupported</c>) when a DWORD came back. The native transport
+    ///     (the default) addresses the right DWORD and decodes the bit.
+    /// </summary>
+    private readonly bool _boolArrayElement;
+
     /// <summary>Builds the (uninitialized) libplctag tag for <paramref name="p"/>.</summary>
     /// <param name="p">The tag creation parameters.</param>
     /// <returns>The new, uninitialised libplctag tag.</returns>
@@ -120,7 +131,10 @@ internal sealed class LibplctagTagRuntime : IAbCipTagRuntime
     {
         AbCipDataType.Bool => bitIndex is int bit
             ? _tag.GetBit(bit)
-            : _tag.GetInt8(offset) != 0,
+            : _boolArrayElement && offset == 0 && _tag.GetSize() == 4
+                ? throw new AbTransportException(AbStatus.Unsupported,
+                    $"'{_tag.Name}' is a BOOL-array element read as a packed DWORD; libplctag cannot say which DWORD it is — use the native transport.")
+                : _tag.GetInt8(offset) != 0,
         AbCipDataType.SInt => (int)(sbyte)_tag.GetInt8(offset),
         AbCipDataType.USInt => (int)_tag.GetUInt8(offset),
         AbCipDataType.Int => (int)_tag.GetInt16(offset),
@@ -221,6 +235,23 @@ internal sealed class LibplctagTagRuntime : IAbCipTagRuntime
         _ => 4,
     };
 
+    /// <inheritdoc />
+    /// <remarks>
+    ///     The codec's bytes are copied into the buffer the last read sized (<c>elem_count</c> elements), so the write
+    ///     sends exactly the new array — never a stale tail. A buffer smaller than the encoded array is refused
+    ///     (<see cref="AbStatus.TooSmall"/>) before anything is changed.
+    /// </remarks>
+    public void EncodeArray(AbCipDataType type, IReadOnlyList<object?> values) =>
+        LibplctagStatusTranslation.WrapSync(() =>
+        {
+            var bytes = AbCipArrayCodec.Encode(type, values);
+            var size = _tag.GetSize();
+            if (size < bytes.Length)
+                throw new AbTransportException(AbStatus.TooSmall,
+                    $"Array write to '{_tag.Name}': {bytes.Length} byte(s) encoded but the tag buffer holds {size}.");
+            for (var i = 0; i < bytes.Length; i++) _tag.SetUInt8(i, bytes[i]);
+        });
+
     /// <inheritdoc />
     public void EncodeValue(AbCipDataType type, int? bitIndex, object? value) =>
         LibplctagStatusTranslation.WrapSync(() => EncodeValueCore(type, bitIndex, value));

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip/NativeAbCipTagRuntime.cs

@@ -41,6 +41,34 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
     private int? _implicitBit;
     private byte[]? _path;
 
+    // U11: a scalar BOOL authored over a BOOL-array element "Bits[n]" — the bit (n % 32) of DWORD n / 32 when the
+    // array is packed. Null for every other tag.
+    private int? _boolElementBit;
+
+    // Both addressings of such a tag: "Bits[n / 32]" (packed: a Logix BOOL array is stored as 0xD3 DWORDs) and the
+    // original "Bits[n]" (per-element: a family that answers 0xC1). _path is whichever _boolLayout selects.
+    private byte[]? _dwordPath;
+    private byte[]? _elementPath;
+
+    /// <summary>How a BOOL-array element tag's array is laid out, as the controller's REPLY TYPE last showed it.</summary>
+    private enum BoolLayout
+    {
+        /// <summary>No reply type seen yet (or only refusals / comms faults): request DWORD-addressed, keep asking.</summary>
+        Unsettled,
+
+        /// <summary>The controller answered 0xD3: address DWORD n / 32, decode bit n % 32.</summary>
+        Packed,
+
+        /// <summary>The controller answered 0xC1: address element n, decode its byte.</summary>
+        PerElement,
+    }
+
+    // #737 fix round 2 (N1): ONLY a reply type settles (or re-settles) the layout — 0xD3 ⇒ Packed, 0xC1 ⇒ PerElement.
+    // A refusal or comms fault says nothing about the layout and leaves it as it was, so a tag deployed before the
+    // online edit that creates its array starts reading correctly once the array exists. A reply whose type disagrees
+    // with the addressing it was asked under is DISCARDED (never decoded) and re-read under the right addressing.
+    private BoolLayout _boolLayout;
+
     private ConnectionLease? _lease;
     private RequestScheduler? _privateScheduler; // ForTest only — production always uses the lease's
     private LogixTagClient? _client;
@@ -51,6 +79,7 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
     private AbStatus _status = AbStatus.Pending;
 
     private byte[]? _writeBuffer;
+    private ushort _writeElements = 1;
     private string? _pendingString;
     private bool _disposed;
 
@@ -94,16 +123,15 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
     /// <summary>The trailing bit selector split off the tag name (<c>Tag.5</c>), if any.</summary>
     internal int? ImplicitBit => _implicitBit;
 
-    // Performs one read to prove the tag exists: libplctag's InitializeAsync errors for an unknown
-    // tag, and the driver's backoff classification depends on create throwing AbStatus.NotFound.
+    // Performs one read to prove the session works and learn the tag's type.
     //
     // A name the encoder refuses throws AbCipTagConfigurationException before any lease: it is this
     // tag's configuration, not the device, and must not black out the device's other tags. A read the
-    // controller answers with a deterministic refusal (AbStatus.TypeMismatch / AbStatus.Unsupported)
-    // proves the tag exists, so it completes the initialization with that status rather than throwing:
-    // a throw here makes the driver dispose the runtime and re-create it - and, for a sole lease
-    // holder, re-Forward-Open - on every poll (mistype churn). The driver's own read reports the
-    // refusal.
+    // controller answers with a deterministic refusal (AbCipRefusals.IsDeviceRefusal: NotFound, OutOfBounds,
+    // NotAllowed, TooLarge/TooSmall, TypeMismatch, Unsupported) proves the session works, so it completes the
+    // initialization with that status rather than throwing: a throw here makes the driver dispose the runtime
+    // and re-create it - and, for a sole lease holder, re-Forward-Open - on every poll (a renamed tag or a
+    // mistype churned the lease). The driver's own read reports the refusal, per tag, on the kept runtime.
     /// <inheritdoc />
     public async Task InitializeAsync(CancellationToken cancellationToken)
     {
@@ -116,7 +144,20 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
             try
             {
                 split = CipPath.SplitBitSelector(_p.TagName);
-                path = CipPath.EncodeSymbolic(split.Name);
+                var name = split.Name;
+                if (split.Bit is null && !_p.IsArray && _p.AuthoredType == AbCipDataType.Bool
+                    && TrySplitBoolArrayElement(name, out var dwordName, out var bit))
+                {
+                    _elementPath = CipPath.EncodeSymbolic(name);
+                    _dwordPath = CipPath.EncodeSymbolic(dwordName);
+                    _boolElementBit = bit;
+                    _boolLayout = BoolLayout.Unsettled;
+                    path = _dwordPath;
+                }
+                else
+                {
+                    path = CipPath.EncodeSymbolic(name);
+                }
             }
             catch (Exception ex) when (ex is ArgumentException or FormatException or OverflowException)
             {
@@ -137,20 +178,45 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
         {
             await ReadAsync(cancellationToken).ConfigureAwait(false);
         }
-        catch (AbTransportException ex) when (ex.Status is AbStatus.TypeMismatch or AbStatus.Unsupported)
+        catch (AbTransportException ex) when (AbCipRefusals.IsDeviceRefusal(ex))
         {
-            // The controller answered — the tag exists. _status already holds the refusal.
+            // The controller answered. _status already holds the refusal.
         }
     }
 
     /// <inheritdoc />
-    public async Task ReadAsync(CancellationToken cancellationToken)
+    public Task ReadAsync(CancellationToken cancellationToken) => ReadCoreAsync(cancellationToken, mayRelayout: true);
+
+    private async Task ReadCoreAsync(CancellationToken cancellationToken, bool mayRelayout)
     {
         var client = Client();
         var path = Path();
         try
         {
             var (type, handle, data) = await client.ReadAsync(path, _elements, cancellationToken, ExpectedElementBytes(_p.AuthoredType)).ConfigureAwait(false);
+            if (_boolElementBit is not null)
+            {
+                BoolLayout? shown = type == LogixTypeCode.BoolArray ? BoolLayout.Packed
+                    : type == LogixTypeCode.Bool ? BoolLayout.PerElement
+                    : null; // any other type (a mistyped tag): no layout evidence; the type check below reports it
+                if (shown is { } layout)
+                {
+                    var askedPacked = ReferenceEquals(path, _dwordPath);
+                    SetBoolLayout(layout);
+                    if ((layout == BoolLayout.Packed) != askedPacked)
+                    {
+                        // The reply is for the OTHER addressing (element n / 32 of a per-element array, or DWORD n of a
+                        // packed one) — not this element. Discard it and ask again under the layout it just showed.
+                        if (mayRelayout)
+                        {
+                            await ReadCoreAsync(cancellationToken, mayRelayout: false).ConfigureAwait(false);
+                            return;
+                        }
+                        throw new AbTransportException(AbStatus.TypeMismatch,
+                            $"'{_p.TagName}': the controller's BOOL-array layout changed between two reads; retrying next read.");
+                    }
+                }
+            }
             _data = data;
             _typeCode = type;
             _structHandle = handle;
@@ -164,6 +230,13 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
         }
     }
 
+    /// <summary>Settles a BOOL-array element tag's layout and selects the matching request path.</summary>
+    private void SetBoolLayout(BoolLayout layout)
+    {
+        _boolLayout = layout;
+        _path = layout == BoolLayout.PerElement ? _elementPath : _dwordPath;
+    }
+
     /// <summary>
     ///     Upper bound on one element's wire bytes for the scheduler's reply budget; 0 when the
     ///     authored type is unknown or a structure, which sends the read alone rather than guess.
@@ -196,6 +269,14 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
 
         try
         {
+            // An unsettled BOOL-array element layout is settled by a read first, so a write never goes to the
+            // DWORD-addressed element of an element-addressed array (#737 I1). A packed layout's write is refused below
+            // (0xD3 DWORD — it would clobber 31 neighbours); a per-element one writes "Bits[n]" as a BOOL.
+            if (_boolElementBit is not null && _boolLayout == BoolLayout.Unsettled)
+            {
+                await ReadAsync(cancellationToken).ConfigureAwait(false);
+                path = Path();
+            }
             if (_pendingString is { } text)
             {
                 // A STRING writes as the type the controller reported — a Logix STRING is a structure
@@ -204,7 +285,7 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
                 if (_typeCode is null || (_typeCode == LogixTypeCode.Struct && _structHandle is null))
                     await ReadAsync(cancellationToken).ConfigureAwait(false);
                 if (_typeCode == LogixTypeCode.Struct && _structHandle is null)
-                    throw new AbTransportException(AbStatus.Unsupported,
+                    throw new AbCipLocalRefusalException(
                         $"STRING write to '{_p.TagName}': the controller returned no structure handle.");
                 var payload = EncodeStringPayload(text);
                 await client.WriteAsync(path, _typeCode!.Value, _structHandle, 1, payload, cancellationToken)
@@ -213,7 +294,7 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
             else if (_writeBuffer is { } buffer)
             {
                 var type = _typeCode ?? AtomicCode(_p.AuthoredType)
-                    ?? throw new AbTransportException(AbStatus.Unsupported,
+                    ?? throw new AbCipLocalRefusalException(
                         $"Write to '{_p.TagName}': type unknown — no prior read and no authored atomic type.");
                 // The read rule decides first: a same-width mismatch (REAL under a DINT-authored tag)
                 // would otherwise send the int bit pattern as a REAL and report Good.
@@ -223,17 +304,18 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
                 // Width is checked only for codes with a known width: DATE / DATE_AND_TIME (0xCD/0xCF) and
                 // Micro800 BYTE/WORD/LWORD (0xD1/0xD2/0xD4) are outside LogixTypeCode's table and write as today.
                 var width = LogixTypeCode.ByteWidth(type);
-                if (width > 0 && width != buffer.Length)
+                if (width > 0 && width * _writeElements != buffer.Length)
                 {
                     // An authored BOOL over a packed BOOL-array DWORD: writing the DWORD would clobber the
                     // 31 sibling bools (libplctag does exactly that), so refuse it as a config error.
-                    if (type == LogixTypeCode.BoolArray)
+                    if (type == LogixTypeCode.BoolArray && _writeElements == 1)
                         throw new NotSupportedException(
                             $"Write to '{_p.TagName}': a BOOL-array element (type 0xD3 DWORD) cannot be written as a single BOOL.");
                     throw new AbTransportException(AbStatus.TypeMismatch,
-                        $"Write to '{_p.TagName}': encoded {buffer.Length} byte(s) but the controller's type 0x{type:X4} is {width} byte(s) wide.");
+                        $"Write to '{_p.TagName}': encoded {buffer.Length} byte(s) for {_writeElements} element(s) but the controller's type 0x{type:X4} is {width} byte(s) wide.");
                 }
-                await client.WriteAsync(path, type, null, 1, buffer, cancellationToken).ConfigureAwait(false);
+                // U10: an array write is ONE Write Tag of _writeElements elements (the client fragments it if large).
+                await client.WriteAsync(path, type, null, _writeElements, buffer, cancellationToken).ConfigureAwait(false);
             }
             else
             {
@@ -276,7 +358,7 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
 
         return type switch
         {
-            AbCipDataType.Bool => bitIndex is int bit ? GetBit(bit) : Span(offset, 1)[0] != 0,
+            AbCipDataType.Bool => bitIndex is int bit ? GetBit(bit) : DecodeScalarBool(offset),
             AbCipDataType.SInt => (int)(sbyte)Span(offset, 1)[0],
             AbCipDataType.USInt => (int)Span(offset, 1)[0],
             AbCipDataType.Int => (int)BinaryPrimitives.ReadInt16LittleEndian(Span(offset, 2)),
@@ -315,10 +397,22 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
         };
     }
 
+    /// <inheritdoc />
+    public void EncodeArray(AbCipDataType type, IReadOnlyList<object?> values)
+    {
+        ArgumentNullException.ThrowIfNull(values);
+        _writeBuffer = null;
+        _pendingString = null;
+        var buffer = AbCipArrayCodec.Encode(type, values); // throws (nothing staged) before any state changes
+        _writeElements = (ushort)Math.Clamp(values.Count, 1, ushort.MaxValue);
+        _writeBuffer = buffer;
+    }
+
     /// <inheritdoc />
     public void EncodeValue(AbCipDataType type, int? bitIndex, object? value)
     {
         _writeBuffer = null;
+        _writeElements = 1;
         _pendingString = null;
         switch (type)
         {
@@ -410,6 +504,47 @@ internal sealed class NativeAbCipTagRuntime : IAbCipTagRuntime
         catch (Exception) { /* ForTest only; Dispose must not throw */ }
     }
 
+    /// <summary>
+    ///     A scalar BOOL. A packed (0xD3) reply is decoded ONLY as the element's own bit of the DWORD the packed layout
+    ///     addressed; any other 0xD3 reply (no element bit, or not settled packed) is refused rather than decoded as
+    ///     "byte 0 != 0" — that value would be some other bit, published as Good (#737 N1).
+    /// </summary>
+    private bool DecodeScalarBool(int offset)
+    {
+        if (offset == 0 && _typeCode == LogixTypeCode.BoolArray)
+        {
+            if (_boolElementBit is int element && _boolLayout == BoolLayout.Packed) return GetBit(element);
+            throw new AbTransportException(AbStatus.TypeMismatch,
+                $"'{_p.TagName}' is a scalar BOOL but the controller answered a packed BOOL-array DWORD (0xD3) with no element bit to select.");
+        }
+        return Span(offset, 1)[0] != 0;
+    }
+
+    /// <summary>
+    ///     Splits a BOOL-array element reference <c>Name[n]</c> (one non-negative subscript, last segment) into the
+    ///     DWORD it lives in (<c>Name[n / 32]</c>) and its bit within that DWORD (<c>n % 32</c>). A Logix BOOL array
+    ///     is stored as DWORDs (type 0xD3) and its element segment addresses those DWORDs.
+    /// </summary>
+    /// <param name="name">The tag name, bit selector already split off.</param>
+    /// <param name="dwordName">The name addressing the containing DWORD.</param>
+    /// <param name="bit">The element's bit within that DWORD.</param>
+    /// <returns><see langword="true"/> when <paramref name="name"/> ends in a single subscript.</returns>
+    internal static bool TrySplitBoolArrayElement(string name, out string dwordName, out int bit)
+    {
+        dwordName = name;
+        bit = 0;
+        if (!name.EndsWith(']')) return false;
+        var open = name.LastIndexOf('[');
+        if (open <= 0) return false;
+        var inner = name[(open + 1)..^1];
+        if (inner.Length == 0 || !inner.All(char.IsAsciiDigit)
+            || !int.TryParse(inner, System.Globalization.NumberStyles.None, System.Globalization.CultureInfo.InvariantCulture, out var n))
+            return false;
+        dwordName = $"{name[..open]}[{n / 32}]";
+        bit = n % 32;
+        return true;
+    }
+
     private byte[] Path() =>
         _path ?? throw new InvalidOperationException("InitializeAsync has not been called.");
 

src/Drivers/ZB.MOM.WW.OtOpcUa.Driver.Calculation/CalculationDriverFactoryExtensions.cs

@@ -38,13 +38,16 @@ public static class CalculationDriverFactoryExtensions
     /// <param name="registry">The driver factory registry to register with.</param>
     /// <param name="loggerFactory">Optional logger factory for per-instance diagnostics loggers.</param>
     /// <param name="scriptRoot">Optional root script logger (fans failure entries onto the <c>script-logs</c> topic).</param>
+    /// <param name="scripting">The node-wide <c>Scripting</c> options every Calculation driver inherits; a key
+    /// authored in a driver's own config overrides that one knob. Null ⇒ the compiled defaults.</param>
     public static void Register(
         DriverFactoryRegistry registry,
         ILoggerFactory? loggerFactory = null,
-        ScriptRootLogger? scriptRoot = null)
+        ScriptRootLogger? scriptRoot = null,
+        ScriptingOptions? scripting = null)
     {
         ArgumentNullException.ThrowIfNull(registry);
-        registry.Register(DriverTypeName, (id, json) => CreateInstance(id, json, loggerFactory, scriptRoot));
+        registry.Register(DriverTypeName, (id, json) => CreateInstance(id, json, loggerFactory, scriptRoot, scripting));
     }
 
     /// <summary>Public overload for the Server-side bootstrapper + test consumers.</summary>
@@ -63,6 +66,18 @@ public static class CalculationDriverFactoryExtensions
     public static CalculationDriver CreateInstance(
         string driverInstanceId, string driverConfigJson,
         ILoggerFactory? loggerFactory, ScriptRootLogger? scriptRoot)
+        => CreateInstance(driverInstanceId, driverConfigJson, loggerFactory, scriptRoot, nodeScripting: null);
+
+    /// <summary>The full overload: also takes the node-wide <c>Scripting</c> options the driver inherits.</summary>
+    /// <param name="driverInstanceId">The unique identifier for the driver instance.</param>
+    /// <param name="driverConfigJson">The merged driver config JSON (carries the <c>RawTags</c> array).</param>
+    /// <param name="loggerFactory">Optional logger factory for per-instance diagnostics loggers.</param>
+    /// <param name="scriptRoot">Optional root script logger for failure fan-out.</param>
+    /// <param name="nodeScripting">The node-wide run-guard options; null ⇒ the compiled defaults.</param>
+    /// <returns>The constructed <see cref="CalculationDriver"/>.</returns>
+    public static CalculationDriver CreateInstance(
+        string driverInstanceId, string driverConfigJson,
+        ILoggerFactory? loggerFactory, ScriptRootLogger? scriptRoot, ScriptingOptions? nodeScripting)
     {
         ArgumentException.ThrowIfNullOrWhiteSpace(driverInstanceId);
         ArgumentException.ThrowIfNullOrWhiteSpace(driverConfigJson);
@@ -71,15 +86,14 @@ public static class CalculationDriverFactoryExtensions
             ?? throw new InvalidOperationException(
                 $"Calculation driver config for '{driverInstanceId}' deserialised to null");
 
-        // Run-guard knobs are optional and default to the node-wide ScriptingOptions defaults; only an
-        // explicitly authored key overrides one (so a config carrying neither binds Scripting = null).
-        ScriptingOptions? scripting = null;
-        if (dto.MaxAllocBytesPerRun is not null || dto.QuarantineAfterConsecutiveTimeouts is not null)
-        {
-            scripting = new ScriptingOptions();
-            if (dto.MaxAllocBytesPerRun is { } alloc) scripting.MaxAllocBytesPerRun = alloc;
-            if (dto.QuarantineAfterConsecutiveTimeouts is { } n) scripting.QuarantineAfterConsecutiveTimeouts = n;
-        }
+        var scripting = ResolveScripting(
+            dto.MaxAllocBytesPerRun, dto.QuarantineAfterConsecutiveTimeouts, nodeScripting, out var adjustments);
+        var logger = loggerFactory?.CreateLogger<CalculationDriver>();
+        // An authored value is never changed silently: the run guard would clamp it anyway, without a word.
+        foreach (var adjustment in adjustments)
+            logger?.LogWarning(
+                "Calculation driver {DriverInstanceId}: its run-guard config was out of range and is clamped: {Adjustment}",
+                driverInstanceId, adjustment);
 
         var options = new CalculationDriverOptions
         {
@@ -89,9 +103,37 @@ public static class CalculationDriverFactoryExtensions
             Scripting = scripting,
         };
 
-        return new CalculationDriver(
-            options, driverInstanceId, scriptRoot,
-            logger: loggerFactory?.CreateLogger<CalculationDriver>());
+        return new CalculationDriver(options, driverInstanceId, scriptRoot, logger: logger);
+    }
+
+    /// <summary>
+    ///     The run-guard options for one Calculation driver. The node-wide options are the base. A key authored
+    ///     in the driver's config overrides only its own knob, on a copy, so the shared node object is never
+    ///     written. With neither, the result is null and the guard uses its compiled defaults.
+    /// </summary>
+    /// <param name="maxAllocBytesPerRun">The driver config's <c>maxAllocBytesPerRun</c>, or null.</param>
+    /// <param name="quarantineAfterConsecutiveTimeouts">The driver config's <c>quarantineAfterConsecutiveTimeouts</c>, or null.</param>
+    /// <param name="nodeScripting">The node-wide options, or null.</param>
+    /// <param name="adjustments">The clamps applied to the driver's OWN keys (the node options were already
+    /// clamped and logged at bind), for the caller to log. Empty when nothing was out of range.</param>
+    /// <returns>The options to hand the driver, or null.</returns>
+    internal static ScriptingOptions? ResolveScripting(
+        long? maxAllocBytesPerRun, int? quarantineAfterConsecutiveTimeouts, ScriptingOptions? nodeScripting,
+        out IReadOnlyList<string> adjustments)
+    {
+        adjustments = [];
+        if (maxAllocBytesPerRun is null && quarantineAfterConsecutiveTimeouts is null) return nodeScripting;
+
+        var resolved = new ScriptingOptions();
+        if (nodeScripting is not null)
+        {
+            resolved.MaxAllocBytesPerRun = nodeScripting.MaxAllocBytesPerRun;
+            resolved.QuarantineAfterConsecutiveTimeouts = nodeScripting.QuarantineAfterConsecutiveTimeouts;
+            resolved.StuckOrphanQuarantineFactor = nodeScripting.StuckOrphanQuarantineFactor;
+        }
+        if (maxAllocBytesPerRun is { } alloc) resolved.MaxAllocBytesPerRun = alloc;
+        if (quarantineAfterConsecutiveTimeouts is { } n) resolved.QuarantineAfterConsecutiveTimeouts = n;
+        return resolved.Sanitized(out adjustments);
     }
 
     /// <summary>The merged-config DTO the calc factory binds from. The Calculation driver has no

src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/Components/Shared/Drivers/Forms/AbCipDriverForm.razor

@@ -21,7 +21,7 @@
             <div class="col-md-3">
                 <label class="form-label">Max concurrent reads</label>
                 <InputNumber @bind-Value="_form.MaxConcurrentReads" @bind-Value:after="EmitAsync" class="form-control form-control-sm" />
-                <div class="form-text">Read work items in flight at once. Default 8; 1 = sequential.</div>
+                <div class="form-text">Read work items in flight at once. Default 8; 1 = sequential (libplctag). The native transport always uses at least 64 — it packs concurrent reads into one request.</div>
             </div>
             <div class="col-md-3">
                 <div class="form-check form-switch mt-4">

src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/ScriptAnalysis/IScriptTagCatalog.cs

@@ -50,7 +50,7 @@ public sealed record ScriptTagInfo(string Path, string Kind, string DataType, st
 ///         else. Verified chain: <c>AddressSpaceComposer</c> harvests the <c>ctx.GetTag("…")</c> literals
 ///         (via <c>EquipmentScriptPaths.ExtractDependencyRefs</c>) into <c>DependencyRefs</c>; those become
 ///         <c>VirtualTagActor._dependencyRefs</c>, registered with <c>DependencyMuxActor</c>, whose
-///         <c>_byRef</c> map is keyed (Ordinal) by <c>DriverInstanceActor.AttributeValuePublished.FullReference</c>
+///         <c>_byRef</c> map is keyed (Ordinal) by <c>DriverInstanceActor.AttributeValueItem.FullReference</c>
 ///         — and in v3 a driver's wire reference <b>is the RawPath</b>
 ///         (<c>DriverHostActor._nodeIdByDriverRef</c> is keyed <c>(DriverInstanceId, RawPath)</c>).
 ///     </para>

src/Server/ZB.MOM.WW.OtOpcUa.AdminUI/ScriptAnalysis/ScriptAnalysisService.cs

@@ -268,7 +268,16 @@ public sealed class ScriptAnalysisService
         var typeInfo = model.GetTypeInfo(memberAccess.Expression);
         var type = typeInfo.Type ?? typeInfo.ConvertedType;
         if (type == null) return null;
-        return type.GetMembers()
+        // The type's own members, then the INSTANCE members it inherits (stopping before System.Object, whose
+        // ToString/GetHashCode noise was never offered). Without the base chain, a member declared only on
+        // ScriptContext — ctx.CancellationToken — compiled and published but never completed. Inherited statics
+        // (ScriptContext.Deadband) are left out: they are not reachable through an instance receiver. This applies to
+        // EVERY receiver, not only ctx: a struct or enum receiver now also offers the System.ValueType /
+        // System.Enum instance members it inherits (CompareTo, HasFlag, ...), all of which compile.
+        var inherited = new List<ISymbol>();
+        for (var b = type.BaseType; b is not null && b.SpecialType != SpecialType.System_Object; b = b.BaseType)
+            inherited.AddRange(b.GetMembers().Where(m => !m.IsStatic));
+        return type.GetMembers().Concat(inherited)
             .Where(m => m.CanBeReferencedByName && !m.IsImplicitlyDeclared)
             // NotApplicable covers members (e.g. some BCL/special members) that carry no explicit accessibility but are referenceable.
             .Where(m => m.DeclaredAccessibility == Accessibility.Public || m.DeclaredAccessibility == Accessibility.NotApplicable)

src/Server/ZB.MOM.WW.OtOpcUa.Host/Configuration/ScriptingOptionsBinding.cs

@@ -0,0 +1,23 @@
+using Microsoft.Extensions.Configuration;
+using ZB.MOM.WW.OtOpcUa.Core.Scripting;
+
+namespace ZB.MOM.WW.OtOpcUa.Host.Configuration;
+
+/// <summary>
+///     Binds the node-wide <c>Scripting</c> section into <see cref="ScriptingOptions"/> and clamps it
+///     (<see cref="ScriptingOptions.Sanitized"/>). One place, so <c>Program.cs</c> and its unit test bind the same
+///     way and a misspelled key cannot silently fall back to the compiled defaults unnoticed.
+/// </summary>
+public static class ScriptingOptionsBinding
+{
+    /// <summary>Binds and clamps the <c>Scripting</c> section.</summary>
+    /// <param name="configuration">The host configuration.</param>
+    /// <param name="adjustments">One line per clamped value, for the caller to log as a Warning.</param>
+    /// <returns>The clamped options; the compiled defaults when the section is absent.</returns>
+    public static ScriptingOptions Bind(IConfiguration configuration, out IReadOnlyList<string> adjustments)
+    {
+        ArgumentNullException.ThrowIfNull(configuration);
+        var bound = configuration.GetSection(ScriptingOptions.SectionName).Get<ScriptingOptions>() ?? new ScriptingOptions();
+        return bound.Sanitized(out adjustments);
+    }
+}

src/Server/ZB.MOM.WW.OtOpcUa.Host/Drivers/DriverFactoryBootstrap.cs

@@ -68,6 +68,8 @@ public static class DriverFactoryBootstrap
             // The calc driver needs the root script logger so a script failure fans out onto the
             // script-logs topic; resolve it here (null on nodes without the script pipeline wired).
             var scriptRoot = sp.GetService<ZB.MOM.WW.OtOpcUa.Core.Scripting.ScriptRootLogger>();
+            // The node-wide Scripting section (bound + clamped in Program.cs); null in test graphs ⇒ defaults.
+            var scripting = sp.GetService<ZB.MOM.WW.OtOpcUa.Core.Scripting.ScriptingOptions>();
             // The Galaxy driver resolves its Gateway.ApiKeySecretRef secret: arm through the
             // shared ISecretResolver (registered unconditionally by AddZbSecrets on every node).
             var secretResolver = sp.GetRequiredService<ISecretResolver>();
@@ -84,7 +86,7 @@ public static class DriverFactoryBootstrap
             // failures; without this it writes to NullLogger in production.
             ConfigureAbTransportLogging(sp);
             Register(registry, loggerFactory, secretResolver, primaryRoleSource, ResolveAbCipDefaultTransport(sp),
-                scriptRoot);
+                scriptRoot, scripting);
             return registry;
         });
         services.AddSingleton<IDriverFactory>(sp =>
@@ -209,11 +211,12 @@ public static class DriverFactoryBootstrap
         ISecretResolver secretResolver,
         IDriverPrimaryRoleSource primaryRoleSource,
         Driver.AbCip.AbCipTransport abCipDefaultTransport,
-        ZB.MOM.WW.OtOpcUa.Core.Scripting.ScriptRootLogger? scriptRoot = null)
+        ZB.MOM.WW.OtOpcUa.Core.Scripting.ScriptRootLogger? scriptRoot = null,
+        ZB.MOM.WW.OtOpcUa.Core.Scripting.ScriptingOptions? scripting = null)
     {
         Driver.AbCip.AbCipDriverFactoryExtensions.Register(registry, loggerFactory, abCipDefaultTransport);
         Driver.AbLegacy.AbLegacyDriverFactoryExtensions.Register(registry, loggerFactory);
-        Driver.Calculation.CalculationDriverFactoryExtensions.Register(registry, loggerFactory, scriptRoot);
+        Driver.Calculation.CalculationDriverFactoryExtensions.Register(registry, loggerFactory, scriptRoot, scripting);
         Driver.FOCAS.FocasDriverFactoryExtensions.Register(registry, loggerFactory);
         Driver.Galaxy.GalaxyDriverFactoryExtensions.Register(registry, secretResolver, loggerFactory);
         Driver.Modbus.ModbusDriverFactoryExtensions.Register(registry, loggerFactory);

src/Server/ZB.MOM.WW.OtOpcUa.Host/Program.cs

@@ -365,10 +365,22 @@ if (hasDriver)
     // F8b — production virtual-tag evaluator (Roslyn-compiled scripts cached per expression).
     // Replaces the F8-default NullVirtualTagEvaluator so VirtualTagActor evaluates real user
     // scripts at runtime.
+    // Node-wide script run-guard knobs (Scripting:MaxAllocBytesPerRun / QuarantineAfterConsecutiveTimeouts /
+    // StuckOrphanQuarantineFactor). Bound once and clamped into sane ranges here, so an absurd value is a startup
+    // Warning rather than an OverflowException inside a script run. The one instance is shared by every script
+    // seam on the node: the virtual-tag evaluator below, the scripted-alarm engine (DriverHostActor, via
+    // WithOtOpcUaRuntimeActors) and every Calculation driver (DriverFactoryBootstrap).
+    var scriptingOptions = ZB.MOM.WW.OtOpcUa.Host.Configuration.ScriptingOptionsBinding.Bind(
+        builder.Configuration, out var scriptingAdjustments);
+    foreach (var adjustment in scriptingAdjustments)
+        Log.Warning("Scripting misconfiguration detected at startup: {Adjustment}", adjustment);
+    builder.Services.AddSingleton(scriptingOptions);
+
     builder.Services.AddSingleton<RoslynVirtualTagEvaluator>(sp =>
         new RoslynVirtualTagEvaluator(
             sp.GetRequiredService<ILoggerFactory>().CreateLogger<RoslynVirtualTagEvaluator>(),
-            sp.GetRequiredService<ScriptRootLogger>()));
+            sp.GetRequiredService<ScriptRootLogger>(),
+            scripting: sp.GetRequiredService<ScriptingOptions>()));
     builder.Services.AddSingleton<IVirtualTagEvaluator>(sp => sp.GetRequiredService<RoslynVirtualTagEvaluator>());
 
     // Script-log fan-out (Layer 0). The DPS publisher resolves the ActorSystem lazily so it never

src/Server/ZB.MOM.WW.OtOpcUa.Runtime/Drivers/DriverHostActor.cs

@@ -54,7 +54,7 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
 /// and the ONLY value ingress from driver children (<see cref="DriverInstanceActor"/> wraps even a
 /// single data change in an <c>AttributeValueBatchPublished</c>). <see cref="FanCommsLossToDependencyMux"/>
 /// fabricates a batch of the same shape straight to the dependency mux, so the host neither receives nor
-/// sends the per-value <c>AttributeValuePublished</c> any more.
+/// sends a per-value message any more (the per-value record is deleted).
 /// </summary>
 public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedStash
 {
@@ -202,6 +202,20 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     private readonly IHistoryWriter _historyWriter;
     private readonly IActorRef? _virtualTagHostOverride;
     private readonly ScriptRootLogger? _scriptRootLogger;
+
+    /// <summary>Node-wide script run-guard options for the scripted-alarm engine; null ⇒ compiled defaults.</summary>
+    private readonly ScriptingOptions? _scripting;
+
+    /// <summary>
+    ///     Driver children this host has stopped (<see cref="StopChild"/>, an operator Restart) and is watching
+    ///     until they terminate. A value burst from one of them is DROPPED in <see cref="ForwardBatchToMux"/>:
+    ///     <c>Context.Stop</c> is asynchronous, so the old child keeps publishing for the rest of the apply
+    ///     handler and its own stop, and forwarding those bursts after <see cref="EvictRetainedValues"/> would
+    ///     re-retain a removed driver's last Good in the dependency mux indefinitely. Akka delivers
+    ///     <see cref="Terminated"/> after the last message the watched child sent here, so removing the entry
+    ///     there bounds the set without ever dropping a live incarnation's values.
+    /// </summary>
+    private readonly HashSet<IActorRef> _retiredChildren = new();
     private readonly IActorRef? _scriptedAlarmHostOverride;
     private readonly ILoggingAdapter _log = Context.GetLogger();
 
@@ -264,6 +278,11 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     /// from <see cref="PushDesiredSubscriptions"/>.</summary>
     private IActorRef? _scriptedAlarmHost;
 
+    /// <summary>The scripted-alarm plans last sent to <see cref="_scriptedAlarmHost"/>, re-sent when it reports a
+    /// supervision restart (<see cref="ScriptedAlarmHostActor.ScriptedAlarmHostRestarted"/>). Null until the
+    /// first apply.</summary>
+    private IReadOnlyList<EquipmentScriptedAlarmPlan>? _lastScriptedAlarmPlans;
+
     private RevisionHash? _currentRevision;
 
     /// <summary>
@@ -641,8 +660,6 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     /// VirtualTag host instead of spawning a real <see cref="VirtualTagHostActor"/> child, so tests
     /// can intercept the <see cref="VirtualTagHostActor.ApplyVirtualTags"/> message. Null in
     /// production (the real host is spawned).</param>
-    /// <param name="loggerFactory">Retained for constructor-signature stability; unused since the
-    /// <c>EfAlarmConditionStateStore</c> fallback that consumed it was retired. Defaults to null.</param>
     /// <param name="scriptRootLogger">Optional root script logger required to spawn the ScriptedAlarm
     /// host (the engine + its script logging hang off it). When null the ScriptedAlarm host is left
     /// unspawned — the graceful dev/None-deployment path.</param>
@@ -694,6 +711,8 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     /// capture to a mesh-partner Primary the cluster failure detector reports unreachable.</param>
     /// <param name="historyUpdateRoleView">Where the OPC UA HistoryUpdate verdict is published for the node manager's
     /// HistoryUpdate Primary gate. Null on nodes that wire no redundancy manager.</param>
+    /// <param name="scripting">The node-wide <c>Scripting</c> run-guard options handed to the scripted-alarm
+    /// engine. Null ⇒ the compiled defaults.</param>
     /// <returns>The Akka.NET <see cref="Akka.Actor.Props"/> used to spawn this actor.</returns>
     public static Props Props(
         IDbContextFactory<OtOpcUaConfigDbContext>? dbFactory,
@@ -707,7 +726,6 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         IVirtualTagEvaluator? virtualTagEvaluator = null,
         IHistoryWriter? historyWriter = null,
         IActorRef? virtualTagHostOverride = null,
-        ILoggerFactory? loggerFactory = null,
         ScriptRootLogger? scriptRootLogger = null,
         IActorRef? scriptedAlarmHostOverride = null,
         IDriverCapabilityInvokerFactory? invokerFactory = null,
@@ -728,7 +746,8 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         Func<CommonsNodeId, bool>? sharesAlarmQueueWithLocal = null,
         TimeSpan? applyOutcomeTimeout = null,
         bool captureWhilePrimaryUnreachable = true,
-        IHistoryUpdateRoleView? historyUpdateRoleView = null) =>
+        IHistoryUpdateRoleView? historyUpdateRoleView = null,
+        ScriptingOptions? scripting = null) =>
         // WARNING: this forwarding list is POSITIONAL, and Props.Create compiles it into an
         // expression tree. Six IActorRef? parameters and several interface-typed ones mean a
         // mis-ordered argument is usually type-compatible and therefore compiles clean, then binds
@@ -737,11 +756,12 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         Akka.Actor.Props.Create(() => new DriverHostActor(
             dbFactory, localNode, ackRouter, driverFactory, localRoles, dependencyMux, opcUaPublishActor,
             healthPublisher, virtualTagEvaluator, historyWriter, virtualTagHostOverride,
-            loggerFactory, scriptRootLogger, scriptedAlarmHostOverride, invokerFactory, driverMemberCountProvider,
+            scriptRootLogger, scriptedAlarmHostOverride, invokerFactory, driverMemberCountProvider,
             deploymentArtifactCache, redundancyRoleView, replicationPeerHost, fetchAndCacheMode, artifactFetcher,
             alarmStateStore, telemetryHub, primaryRoleSource, historizationRoleView,
             sharesHistorizationScopeWithLocal, applyWedgeTimeout, reprovisionBackoff, commsRecoveryAuditDelay,
-            sharesAlarmQueueWithLocal, applyOutcomeTimeout, captureWhilePrimaryUnreachable, historyUpdateRoleView));
+            sharesAlarmQueueWithLocal, applyOutcomeTimeout, captureWhilePrimaryUnreachable, historyUpdateRoleView,
+            scripting));
 
     /// <summary>Initializes a new DriverHostActor with the specified dependencies.</summary>
     /// <param name="dbFactory">Database context factory for configuration database access.</param>
@@ -760,8 +780,6 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     /// for historized VirtualTag results; defaults to <see cref="NullHistoryWriter"/>.</param>
     /// <param name="virtualTagHostOverride">Test seam: when supplied, used as the VirtualTag host
     /// instead of spawning a real <see cref="VirtualTagHostActor"/> child.</param>
-    /// <param name="loggerFactory">Retained for constructor-signature stability; unused since the
-    /// <c>EfAlarmConditionStateStore</c> fallback that consumed it was retired. Defaults to null.</param>
     /// <param name="scriptRootLogger">Optional root script logger required to spawn the ScriptedAlarm
     /// host; when null the host is left unspawned.</param>
     /// <param name="scriptedAlarmHostOverride">Test seam: when supplied, used as the ScriptedAlarm host
@@ -830,6 +848,8 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     /// (<see cref="PrimaryGatePolicy.ShouldServeHistoryUpdate"/> — the data-plane verdict, but closed until the first
     /// redundancy snapshot) is published for the node manager's HistoryUpdate Primary gate. Null on nodes that wire no
     /// redundancy manager.</param>
+    /// <param name="scripting">The node-wide <c>Scripting</c> run-guard options handed to the scripted-alarm
+    /// engine. Null ⇒ the compiled defaults.</param>
     public DriverHostActor(
         IDbContextFactory<OtOpcUaConfigDbContext>? dbFactory,
         CommonsNodeId localNode,
@@ -842,7 +862,6 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         IVirtualTagEvaluator? virtualTagEvaluator = null,
         IHistoryWriter? historyWriter = null,
         IActorRef? virtualTagHostOverride = null,
-        ILoggerFactory? loggerFactory = null,
         ScriptRootLogger? scriptRootLogger = null,
         IActorRef? scriptedAlarmHostOverride = null,
         IDriverCapabilityInvokerFactory? invokerFactory = null,
@@ -863,8 +882,10 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         Func<CommonsNodeId, bool>? sharesAlarmQueueWithLocal = null,
         TimeSpan? applyOutcomeTimeout = null,
         bool captureWhilePrimaryUnreachable = true,
-        IHistoryUpdateRoleView? historyUpdateRoleView = null)
+        IHistoryUpdateRoleView? historyUpdateRoleView = null,
+        ScriptingOptions? scripting = null)
     {
+        _scripting = scripting;
         _captureWhilePrimaryUnreachable = captureWhilePrimaryUnreachable;
         _historyUpdateRoleView = historyUpdateRoleView;
         _applyOutcomeTimeout = applyOutcomeTimeout ?? DefaultApplyOutcomeTimeout;
@@ -1146,8 +1167,9 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     ///     null and ApplyScriptedAlarms becomes a no-op. The engine is built around a fresh
     ///     <see cref="DependencyMuxTagUpstreamSource"/> + the wired <see cref="IAlarmStateStore"/> (the
     ///     replicated <c>LocalDbAlarmConditionStateStore</c> on a driver-role node); when no store was
-    ///     wired the host is skipped. The host (spawned as a child) owns + disposes the engine in its
-    ///     PostStop, so it stops with the driver host.
+    ///     wired the host is skipped. The host is spawned through the factory <c>Props</c>, so EACH
+    ///     incarnation (a supervision restart included) builds, owns and disposes its own engine; it stops
+    ///     with the driver host.
     /// </summary>
     private void SpawnScriptedAlarmHost()
     {
@@ -1178,13 +1200,23 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
             return;
         }
 
+        // Factory Props: every incarnation of the host (a supervision restart included) builds and owns a fresh
+        // (upstream, engine) pair, so a restart never reuses an engine the previous incarnation disposed or left
+        // half-loaded. The closure captures locals only; it runs in the child's constructor.
         var store = _alarmStateStore;
-        var upstream = new DependencyMuxTagUpstreamSource();
-        var engine = new ScriptedAlarmEngine(
-            upstream, store, new ScriptLoggerFactory(_scriptRootLogger.Logger), _scriptRootLogger.Logger);
+        var scriptLogger = _scriptRootLogger.Logger;
+        var scripting = _scripting;
         _scriptedAlarmHost = Context.ActorOf(
             ScriptedAlarmHostActor.Props(
-                _opcUaPublishActor, _dependencyMux, upstream, engine, _localNode, telemetryHub: _telemetryHub),
+                _opcUaPublishActor, _dependencyMux,
+                () =>
+                {
+                    var upstream = new DependencyMuxTagUpstreamSource();
+                    var engine = new ScriptedAlarmEngine(
+                        upstream, store, new ScriptLoggerFactory(scriptLogger), scriptLogger, scripting: scripting);
+                    return (upstream, engine);
+                },
+                _localNode, telemetryHub: _telemetryHub),
             "scripted-alarm-host");
     }
 
@@ -1500,6 +1532,8 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         Receive<ApplyWedged>(HandleApplyWedged);
         Receive<GetDiagnostics>(HandleGetDiagnostics);
         Receive<DriverInstanceActor.AttributeValueBatchPublished>(ForwardBatchToMux);
+        Receive<Terminated>(OnRetiredChildTerminated);
+        Receive<ScriptedAlarmHostActor.ScriptedAlarmHostRestarted>(_ => HandleScriptedAlarmHostRestarted());
         Receive<DriverInstanceActor.AttributeAlarmPublished>(ForwardNativeAlarm);
         Receive<DriverInstanceActor.ConnectivityChanged>(OnDriverConnectivityChanged);
         Receive<CommsRecoveryAudit>(HandleCommsRecoveryAudit);
@@ -1562,6 +1596,8 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         Receive<ApplyWedged>(HandleApplyWedged);
         Receive<GetDiagnostics>(HandleGetDiagnostics);
         Receive<DriverInstanceActor.AttributeValueBatchPublished>(ForwardBatchToMux);
+        Receive<Terminated>(OnRetiredChildTerminated);
+        Receive<ScriptedAlarmHostActor.ScriptedAlarmHostRestarted>(_ => HandleScriptedAlarmHostRestarted());
         Receive<DriverInstanceActor.AttributeAlarmPublished>(ForwardNativeAlarm);
         Receive<DriverInstanceActor.ConnectivityChanged>(OnDriverConnectivityChanged);
         Receive<CommsRecoveryAudit>(HandleCommsRecoveryAudit);
@@ -1606,6 +1642,16 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     /// </summary>
     private void ForwardBatchToMux(DriverInstanceActor.AttributeValueBatchPublished msg)
     {
+        // A burst from an incarnation this host already stopped: forwarding it would undo the SourceRemoved
+        // eviction (see _retiredChildren). Its ConnectivityChanged is NOT filtered — the respawn path relies on
+        // that Bad fan to cover the gap.
+        if (_retiredChildren.Contains(Sender))
+        {
+            _log.Debug("DriverHost {Node}: dropped a {Count}-item value burst from a stopped incarnation of {Id}",
+                _localNode, msg.Items.Count, msg.DriverInstanceId);
+            return;
+        }
+
         // One envelope per burst; the mux filters per item (it registered a batch receive).
         _dependencyMux?.Tell(msg);
 
@@ -2501,6 +2547,8 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         Receive<DriverInstanceActor.AttributeValueBatchPublished>(m =>
             _log.Debug("DriverHost {Node}: dropping a {Count}-item value burst from {Id} while Stale (no address space)",
                 _localNode, m.Items.Count, m.DriverInstanceId));
+        Receive<Terminated>(OnRetiredChildTerminated);
+        Receive<ScriptedAlarmHostActor.ScriptedAlarmHostRestarted>(_ => HandleScriptedAlarmHostRestarted());
         Receive<DriverInstanceActor.AttributeAlarmPublished>(m =>
             _log.Debug("DriverHost {Node}: dropping a native alarm transition ({Cond}) from {Id} while Stale",
                 _localNode, m.Args.ConditionId, m.DriverInstanceId));
@@ -3652,6 +3700,7 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         // (re)loads its engine + re-registers mux interest for the union of dependency refs. Covers
         // both the fresh-apply and bootstrap-restore paths (both call this method); the Stale-recovery
         // path deliberately does not, matching driver + VirtualTag recovery.
+        _lastScriptedAlarmPlans = composition.EquipmentScriptedAlarms;
         _scriptedAlarmHost?.Tell(new ScriptedAlarmHostActor.ApplyScriptedAlarms(composition.EquipmentScriptedAlarms));
         if (composition.EquipmentScriptedAlarms.Count > 0)
         {
@@ -3816,8 +3865,10 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
     {
         if (!_children.TryGetValue(driverInstanceId, out var entry)) return;
         StopDependencyAdapter(driverInstanceId);
+        RetireChild(entry.Actor);
         Context.Stop(entry.Actor);
         _children.Remove(driverInstanceId);
+        EvictRetainedValues(driverInstanceId);
         // The instance left the plan (StopChild is only reached via plan.ToStop), so its cached
         // Health/Resilience snapshots must not replay to new telemetry subscribers. Release the shared
         // resilience state FIRST: the resilience publisher re-emits the tracker's snapshots every tick, so
@@ -3836,6 +3887,56 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
         _log.Info("DriverHost {Node}: stopped driver child {Id}", _localNode, driverInstanceId);
     }
 
+    /// <summary>
+    ///     Tells the dependency mux to drop every retained value this driver instance produced, so a stopped
+    ///     driver's last Good is never replayed to a later replay-opting subscriber as if it were current. Sent
+    ///     wherever a driver OBJECT ends: a deploy removal, the stop half of a config-change respawn
+    ///     (<see cref="StopChild"/>) and an operator Restart (a new driver object). Deliberately NOT sent on a
+    ///     supervision restart (<see cref="HandleReprovisionRequested"/>): that re-uses the same driver, whose last value is
+    ///     still its current one. Sent before any respawn, on the same host→mux pair the new incarnation's
+    ///     values travel, so the eviction is ordered ahead of them.
+    /// </summary>
+    /// <remarks>
+    ///     The stopped child keeps publishing until it actually terminates, and for a REMOVED driver nothing
+    ///     would ever overwrite a value re-retained after this eviction (it never publishes again, and its
+    ///     PostStop Bad fan resolves no refs once the apply has rebuilt the index without it). So the caller
+    ///     also <see cref="RetireChild"/>s it, and <see cref="ForwardBatchToMux"/> drops every burst from a
+    ///     retired child.
+    /// </remarks>
+    /// <param name="driverInstanceId">The driver instance that ended.</param>
+    private void EvictRetainedValues(string driverInstanceId) =>
+        _dependencyMux?.Tell(new DependencyMuxActor.SourceRemoved(driverInstanceId));
+
+    /// <summary>Marks a child this host is about to stop as retired and watches it until it terminates.</summary>
+    /// <param name="child">The child being stopped.</param>
+    private void RetireChild(IActorRef child)
+    {
+        _retiredChildren.Add(child);
+        Context.Watch(child);
+    }
+
+    /// <summary>
+    ///     The scripted-alarm host was restarted by supervision and has an empty engine: re-send the last plans so
+    ///     its inputs and operator commands work again without waiting for the next deploy. Only the CURRENT host's
+    ///     signal is honoured.
+    /// </summary>
+    private void HandleScriptedAlarmHostRestarted()
+    {
+        if (_scriptedAlarmHost is null || !Sender.Equals(_scriptedAlarmHost)) return;
+        if (_lastScriptedAlarmPlans is null)
+        {
+            _log.Debug("DriverHost {Node}: ScriptedAlarm host restarted before any apply; nothing to re-send", _localNode);
+            return;
+        }
+        _log.Warning("DriverHost {Node}: ScriptedAlarm host was restarted by supervision; re-applying {Count} scripted alarm(s)",
+            _localNode, _lastScriptedAlarmPlans.Count);
+        _scriptedAlarmHost.Tell(new ScriptedAlarmHostActor.ApplyScriptedAlarms(_lastScriptedAlarmPlans));
+    }
+
+    /// <summary>A retired child has terminated; nothing more can arrive from it.</summary>
+    /// <param name="msg">The termination notice.</param>
+    private void OnRetiredChildTerminated(Terminated msg) => _retiredChildren.Remove(msg.ActorRef);
+
     private static string ActorNameFor(string driverInstanceId, long generation)
     {
         // Akka actor names cannot contain '/', ':', or whitespace. Mangle defensively. The monotonic
@@ -3881,8 +3982,12 @@ public sealed class DriverHostActor : ReceiveActor, IWithTimers, IWithUnboundedS
 
         // Stop the existing child actor — DriverInstanceActor.PostStop calls ShutdownAsync.
         StopDependencyAdapter(msg.DriverInstanceId);
+        RetireChild(entry.Actor);
         Context.Stop(entry.Actor);
         _children.Remove(msg.DriverInstanceId);
+        // An operator Restart builds a NEW driver object, so the old one's retained values are evicted like a
+        // removal's; the respawned driver republishes its own.
+        EvictRetainedValues(msg.DriverInstanceId);
         // The old incarnation's health must not read as the new one's until it publishes its own.
         _healthPublisher.Forget(msg.DriverInstanceId);
 

src/Server/ZB.MOM.WW.OtOpcUa.Runtime/Drivers/DriverInstanceActor.cs

@@ -108,21 +108,16 @@ public sealed class DriverInstanceActor : ReceiveActor, IWithTimers
     ///     state — a no-op when already Reconnecting or Connecting.
     /// </summary>
     public sealed record ForceReconnect;
-    /// <summary>One driver-published value. <b>No production sender remains</b> — the subscription path
-    /// batches every data change into an <see cref="AttributeValueBatchPublished"/>, and the comms-loss fan
-    /// (<c>DriverHostActor.FanCommsLossToDependencyMux</c>), which was the last sender to fabricate this
-    /// message, now fabricates a batch too. The dependency mux still accepts it for tests and direct senders.
+    /// <summary>One value within an <see cref="AttributeValueBatchPublished"/>.
     /// <para><paramref name="StatusCode"/> is the driver's full OPC UA status code including its sub-code
     /// — not a severity bucket. Classify it with <see cref="OpcUaStatus.IsGood"/> and friends rather
-    /// than comparing it to <see cref="OpcUaStatus.Good"/>.</para></summary>
-    public sealed record AttributeValuePublished(string DriverInstanceId, string FullReference, object? Value, uint StatusCode, DateTime TimestampUtc);
-    /// <summary>One value within an <see cref="AttributeValueBatchPublished"/>.
+    /// than comparing it to <see cref="OpcUaStatus.Good"/>.</para>
     /// <para>A struct deliberately: an <c>AttributeValueItem[]</c> then holds its items inline, so a burst of
     /// N values costs one array allocation instead of one array plus N boxes. Indexed access through
     /// <see cref="IReadOnlyList{T}"/> does not box.</para></summary>
     public readonly record struct AttributeValueItem(string FullReference, object? Value, uint StatusCode, DateTime TimestampUtc);
-    /// <summary>Batched form of <see cref="AttributeValuePublished"/> — every changed attribute of ONE
-    /// driver production burst in one message. The parent fans the whole batch to the mux and the OPC UA
+    /// <summary>The ONLY value message a driver child sends: every changed attribute of ONE driver production
+    /// burst in one message (a single data change is a one-item batch). The parent fans the whole batch to the mux and the OPC UA
     /// publish path in one hop each. Status codes ride verbatim.</summary>
     public sealed record AttributeValueBatchPublished(string DriverInstanceId, IReadOnlyList<AttributeValueItem> Items);
     private sealed record DataChangeForward(string FullReference, DataValueSnapshot Snapshot);
@@ -132,7 +127,7 @@ public sealed class DriverInstanceActor : ReceiveActor, IWithTimers
     private sealed record DataChangeBatchForward(IReadOnlyList<DataChangeItem> Items);
     /// <summary>Published to the parent whenever the subscribed driver (an <see cref="IAlarmSource"/>) fires
     /// <see cref="IAlarmSource.OnAlarmEvent"/>. The parent (<see cref="DriverHostActor"/>) projects + routes it
-    /// to the materialised Part 9 condition. Parallels <see cref="AttributeValuePublished"/>.</summary>
+    /// to the materialised Part 9 condition. Parallels <see cref="AttributeValueBatchPublished"/>.</summary>
     public sealed record AttributeAlarmPublished(string DriverInstanceId, AlarmEventArgs Args);
     private sealed record NativeAlarmRaised(AlarmEventArgs Args);
     /// <summary>Self-sent on Connected entry / when alarm refs are (re)pushed, to establish the native-alarm

src/Server/ZB.MOM.WW.OtOpcUa.Runtime/ScriptedAlarms/ScriptedAlarmHostActor.cs

@@ -86,6 +86,21 @@ public sealed class ScriptedAlarmHostActor : ReceiveActor
     /// <param name="Plans">The desired Equipment-namespace scripted-alarm plans.</param>
     public sealed record ApplyScriptedAlarms(IReadOnlyList<EquipmentScriptedAlarmPlan> Plans);
 
+    /// <summary>
+    ///     Told to the parent from <see cref="PostRestart"/>. A supervision restart is otherwise invisible to the
+    ///     parent, and the new incarnation has an EMPTY engine (factory Props) and no mux interest, so until the
+    ///     next <see cref="ApplyScriptedAlarms"/> its inputs stop and every operator command is dropped as not
+    ///     owned. The parent (<c>DriverHostActor</c>) answers by re-sending the last plans it applied, mirroring
+    ///     how a driver child's <c>ReprovisionRequested</c> is answered.
+    /// </summary>
+    public sealed class ScriptedAlarmHostRestarted
+    {
+        /// <summary>The single instance.</summary>
+        public static readonly ScriptedAlarmHostRestarted Instance = new();
+
+        private ScriptedAlarmHostRestarted() { }
+    }
+
     /// <summary>Marshals an engine emission off the engine's worker thread onto the actor thread.
     /// Carries the <see cref="ScriptedAlarmEvent"/> the engine raised on <c>OnEvent</c>.</summary>
     private sealed record EngineEmission(ScriptedAlarmEvent Event);
@@ -384,9 +399,10 @@ public sealed class ScriptedAlarmHostActor : ReceiveActor
         // subscriptions exist, so any DependencyValueChanged the mux delivers will be observed by the
         // engine. RegisterInterest replaces a subscriber's prior interest set (see DependencyMuxActor),
         // so a re-Apply with a fresh union simply supersedes the old one — no explicit unregister needed.
-        // ReplayRetained: the mux replays the last value it saw for each ref, so a fresh engine (a
-        // supervision restart builds one around an EMPTY upstream) converges immediately instead of
-        // waiting for every input to change again.
+        // ReplayRetained: the mux replays the last value it saw for each ref, so a freshly loaded engine
+        // converges immediately instead of waiting for every input to change again. That includes the engine
+        // a supervision restart builds around an EMPTY upstream, but only once the parent has re-sent
+        // ApplyScriptedAlarms in answer to ScriptedAlarmHostRestarted (PostRestart); nothing reloads it before.
         _mux?.Tell(new DependencyMuxActor.RegisterInterest(msg.DepRefs, Self, ReplayRetained: true));
         // Prune the upstream's retained snapshots to the loaded set (R02:L3): a path no alarm references
         // any more will never be pushed again, so its last value is dead weight for the process lifetime.
@@ -791,6 +807,15 @@ public sealed class ScriptedAlarmHostActor : ReceiveActor
         base.PreRestart(reason, message);
     }
 
+    /// <inheritdoc />
+    /// <remarks>Tells the parent <see cref="ScriptedAlarmHostRestarted"/> so it re-sends the last
+    /// <see cref="ApplyScriptedAlarms"/>: the new incarnation has nothing loaded and no mux interest.</remarks>
+    protected override void PostRestart(Exception reason)
+    {
+        base.PostRestart(reason);
+        Context.Parent.Tell(ScriptedAlarmHostRestarted.Instance);
+    }
+
     /// <inheritdoc />
     protected override void PostStop()
     {

src/Server/ZB.MOM.WW.OtOpcUa.Runtime/ServiceCollectionExtensions.cs

@@ -609,7 +609,6 @@ public static class ServiceCollectionExtensions
                     healthPublisher: healthPublisher,
                     virtualTagEvaluator: virtualTagEvaluator,
                     historyWriter: historyWriter,
-                    loggerFactory: loggerFactory,
                     scriptRootLogger: scriptRootLogger,
                     invokerFactory: invokerFactory,
                     deploymentArtifactCache: deploymentArtifactCache,
@@ -638,7 +637,10 @@ public static class ServiceCollectionExtensions
                     sharesAlarmQueueWithLocal: BuildAlarmQueuePeerPredicate(
                         resolver.GetService<IConfiguration>(), roleInfo, replicationPeerHost),
                     // The HistoryUpdate verdict (closed until the first redundancy snapshot).
-                    historyUpdateRoleView: resolver.GetService<IHistoryUpdateRoleView>()),
+                    historyUpdateRoleView: resolver.GetService<IHistoryUpdateRoleView>(),
+                    // The node-wide Scripting section (bound + clamped in Program.cs) for the scripted-alarm
+                    // engine. Unbound in test graphs ⇒ null ⇒ the compiled defaults.
+                    scripting: resolver.GetService<ScriptingOptions>()),
                 DriverHostActorName);
             registry.Register<DriverHostActorKey>(driverHost);
 

src/Server/ZB.MOM.WW.OtOpcUa.Runtime/VirtualTags/DependencyMuxActor.cs

@@ -15,11 +15,9 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.VirtualTags;
 ///     DriverHostActor forwards driver-child value publishes to this mux (one mux per driver-role
 ///     node). The ONLY driver ingress is the batched
 ///     <see cref="DriverInstanceActor.AttributeValueBatchPublished"/> — one envelope per driver
-///     production burst, which the mux filters per item. The per-value
-///     <see cref="DriverInstanceActor.AttributeValuePublished"/> receive has <b>no production sender left
-///     at all</b>: the comms-loss fan (<c>DriverHostActor.FanCommsLossToDependencyMux</c>) was its last
-///     one and now fabricates an <c>AttributeValueBatchPublished</c> instead. The receive is kept for
-///     tests and direct senders — treat it as dead in production.
+///     production burst, which the mux filters per item. The comms-loss fan
+///     (<c>DriverHostActor.FanCommsLossToDependencyMux</c>) fabricates one too. There is no per-value
+///     ingress: a single value is a one-item batch (tests build one with <c>ValuePublishes.Single</c>).
 ///
 ///     <b>The mux emits ONLY batches</b>, so a subscriber's
 ///     batch receive is its complete hot path. Every subscriber keeps a per-value
@@ -52,11 +50,9 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.VirtualTags;
 ///         <b>Source removal.</b> The mux records which driver instance last published each retained ref;
 ///         <see cref="SourceRemoved"/> evicts every value that driver produced, so a removed driver's last
 ///         Good is never replayed as current. A comms-loss fan (a Bad batch) already overwrites the retained
-///         value with Bad. ⚠️ As of fix round 1 no production actor sends <see cref="SourceRemoved"/> yet —
-///         <c>DriverHostActor</c> (outside the scripting lane) must Tell it when it stops / removes a driver
-///         child. Until then a removed driver's last value can still be replayed to a replay-opting
-///         subscriber that re-registers for that ref while another replay-opting subscriber holds it, or
-///         within the grace window.
+///         value with Bad. <c>DriverHostActor</c> sends <see cref="SourceRemoved"/> wherever a driver object
+///         ends (a deploy removal, the stop half of a config-change respawn, an operator Restart), and never
+///         on a supervision restart, which re-uses the same driver.
 ///     </para>
 /// </summary>
 public sealed class DependencyMuxActor : ReceiveActor, IWithTimers
@@ -154,7 +150,6 @@ public sealed class DependencyMuxActor : ReceiveActor, IWithTimers
         Receive<UnregisterInterest>(msg => RemoveSubscriber(msg.Subscriber));
         Receive<SourceRemoved>(OnSourceRemoved);
         Receive<SweepReleased>(_ => OnSweep());
-        Receive<DriverInstanceActor.AttributeValuePublished>(OnAttributeValuePublished);
         Receive<DriverInstanceActor.AttributeValueBatchPublished>(OnAttributeValueBatchPublished);
         Receive<Terminated>(msg => RemoveSubscriber(msg.ActorRef));
     }
@@ -265,30 +260,6 @@ public sealed class DependencyMuxActor : ReceiveActor, IWithTimers
         _log.Debug("DependencyMux: source {Driver} removed — evicted {Count} retained value(s)", msg.DriverInstanceId, evicted);
     }
 
-    private void OnAttributeValuePublished(DriverInstanceActor.AttributeValuePublished msg)
-    {
-        if (!_byRef.TryGetValue(msg.FullReference, out var entry))
-        {
-            // Nothing is registered for (or retaining) this ref — drop. Common in normal operation; the address
-            // space carries thousands of tags and only a fraction feed virtual-tag expressions.
-            return;
-        }
-        var dep = new VirtualTagActor.DependencyValueChanged(msg.FullReference, msg.Value, msg.TimestampUtc, msg.StatusCode);
-        if (entry.Retains)
-        {
-            entry.Last = dep;
-            entry.LastSource = msg.DriverInstanceId;
-        }
-        foreach (var sub in entry.Subscribers)
-        {
-            // A single-item batch, so the mux emits exactly ONE message shape and every subscriber needs
-            // only the batch receive on its hot path. One list per subscriber: batch ownership transfers
-            // to the receiver, so a shared list must never cross mailboxes. Nothing in production reaches
-            // this path (see the class doc) — the extra allocation is noise on a test-only seam.
-            sub.Tell(new VirtualTagActor.DependencyValueBatch(new[] { dep }));
-        }
-    }
-
     private void OnAttributeValueBatchPublished(DriverInstanceActor.AttributeValueBatchPublished msg)
     {
         // Nothing registered and nothing retained — skip the whole burst rather than probing an empty map per item.

tests/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting.Tests/ScriptCancellationTokenTests.cs

@@ -0,0 +1,67 @@
+using Shouldly;
+using Xunit;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Scripting.Tests;
+
+/// <summary>
+///     <c>ctx.CancellationToken</c> — the run token, visible to scripts. <see cref="ScriptRunGuard"/> already
+///     cancels the token it hands <see cref="ScriptEvaluator{TContext, TResult}.RunAsync"/> when a run times out;
+///     a script that polls it can now end its own abandoned run instead of burning a pool thread until it
+///     returns. Cooperative only: a script that ignores it is still an orphan.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class ScriptCancellationTokenTests
+{
+    /// <summary>Spins until cancelled, but ends by itself after 10 s so a broken implementation cannot pin a
+    /// pool thread past the test.</summary>
+    private const string CooperativeSpin =
+        """
+        var end = System.DateTime.UtcNow.AddSeconds(10);
+        while (!ctx.CancellationToken.IsCancellationRequested && System.DateTime.UtcNow < end) { }
+        return ctx.CancellationToken.IsCancellationRequested ? -1 : 1;
+        """;
+
+    [Fact]
+    public async Task A_script_sees_an_uncancelled_token_on_a_normal_run()
+    {
+        using var eval = ScriptEvaluator<FakeScriptContext, bool>.Compile(
+            "return ctx.CancellationToken.CanBeCanceled && !ctx.CancellationToken.IsCancellationRequested;");
+        using var cts = new CancellationTokenSource();
+
+        (await eval.RunAsync(new FakeScriptContext(), cts.Token)).ShouldBeTrue();
+    }
+
+    [Fact]
+    public async Task Outside_a_run_the_token_is_None()
+    {
+        using var eval = ScriptEvaluator<FakeScriptContext, int>.Compile("return 1;");
+        var ctx = new FakeScriptContext();
+        (await eval.RunAsync(ctx)).ShouldBe(1);
+
+        ctx.CancellationToken.ShouldBe(CancellationToken.None, "the token is scoped to the run, not left on the context");
+    }
+
+    [Fact]
+    public async Task A_timed_out_run_whose_script_polls_the_token_ends_its_own_orphan()
+    {
+        var guard = new ScriptRunGuard();
+        using var eval = ScriptEvaluator<FakeScriptContext, int>.Compile(CooperativeSpin);
+        var ct = TestContext.Current.CancellationToken;
+
+        await Should.ThrowAsync<ScriptTimeoutException>(
+            () => guard.RunAsync("k", CooperativeSpin, eval, new FakeScriptContext(), TimeSpan.FromMilliseconds(100), ct));
+
+        // The guard cancelled the run token on timeout; the script saw it and returned, which frees the slot well
+        // inside its own 10 s fallback. Without ctx.CancellationToken the orphan would hold it for the full 10 s.
+        var deadline = DateTime.UtcNow.AddSeconds(3);
+        while (guard.InFlightCount != 0 && DateTime.UtcNow < deadline)
+            await Task.Delay(20, ct);
+        guard.InFlightCount.ShouldBe(0, "the cooperative script ended its abandoned run");
+
+        // Positive control: the slot is free, so the next run is dispatched rather than refused. Same source
+        // identity on purpose: a different one would reset the key and dispatch even behind a live orphan.
+        using var fast = ScriptEvaluator<FakeScriptContext, int>.Compile("return 5;");
+        (await guard.RunAsync("k", CooperativeSpin, fast, new FakeScriptContext(), TimeSpan.FromSeconds(5), ct))
+            .ShouldBe(5);
+    }
+}

tests/Core/ZB.MOM.WW.OtOpcUa.Core.Scripting.Tests/ScriptingOptionsTests.cs

@@ -0,0 +1,109 @@
+using Shouldly;
+using Xunit;
+
+namespace ZB.MOM.WW.OtOpcUa.Core.Scripting.Tests;
+
+/// <summary>
+///     <see cref="ScriptingOptions.Sanitized"/> — the bind-time clamp of the <c>Scripting</c> section's numeric
+///     knobs. An absurd value must be pulled into a sane range at bind time rather than surfacing later as an
+///     <see cref="OverflowException"/> inside a script run (<c>timeout * StuckOrphanQuarantineFactor</c>).
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class ScriptingOptionsTests
+{
+    [Fact]
+    public void Defaults_pass_through_unchanged_with_no_adjustments()
+    {
+        var sanitized = new ScriptingOptions().Sanitized(out var adjustments);
+
+        adjustments.ShouldBeEmpty();
+        sanitized.MaxAllocBytesPerRun.ShouldBe(ScriptingOptions.DefaultMaxAllocBytesPerRun);
+        sanitized.QuarantineAfterConsecutiveTimeouts.ShouldBe(ScriptingOptions.DefaultQuarantineAfterConsecutiveTimeouts);
+        sanitized.StuckOrphanQuarantineFactor.ShouldBe(ScriptingOptions.DefaultStuckOrphanQuarantineFactor);
+    }
+
+    [Theory]
+    [InlineData(double.MaxValue, ScriptingOptions.MaxStuckOrphanQuarantineFactor)]
+    [InlineData(double.PositiveInfinity, ScriptingOptions.MaxStuckOrphanQuarantineFactor)]
+    [InlineData(1e300, ScriptingOptions.MaxStuckOrphanQuarantineFactor)]
+    [InlineData(double.NaN, ScriptingOptions.DefaultStuckOrphanQuarantineFactor)]
+    [InlineData(-5, 0)]
+    [InlineData(double.NegativeInfinity, 0)]
+    [InlineData(0, 0)]
+    [InlineData(2.5, 2.5)]
+    public void StuckOrphanQuarantineFactor_is_clamped(double configured, double expected)
+    {
+        var sanitized = new ScriptingOptions { StuckOrphanQuarantineFactor = configured }.Sanitized(out var adjustments);
+
+        sanitized.StuckOrphanQuarantineFactor.ShouldBe(expected);
+        // Negative already means "disabled" and is documented as such; normalising it to 0 is not worth a warning.
+        if (!double.IsNaN(configured) && configured >= 0 && configured <= ScriptingOptions.MaxStuckOrphanQuarantineFactor)
+            adjustments.ShouldBeEmpty();
+    }
+
+    [Theory]
+    [InlineData(0, 1)]
+    [InlineData(-3, 1)]
+    [InlineData(int.MaxValue, ScriptingOptions.MaxQuarantineAfterConsecutiveTimeouts)]
+    [InlineData(5, 5)]
+    public void QuarantineAfterConsecutiveTimeouts_is_clamped(int configured, int expected) =>
+        new ScriptingOptions { QuarantineAfterConsecutiveTimeouts = configured }.Sanitized(out _)
+            .QuarantineAfterConsecutiveTimeouts.ShouldBe(expected);
+
+    [Theory]
+    [InlineData(-1L, 0L)]
+    [InlineData(0L, 0L)]
+    [InlineData(1024L, ScriptingOptions.MinMaxAllocBytesPerRun)]
+    [InlineData(long.MaxValue, long.MaxValue)]
+    public void MaxAllocBytesPerRun_is_clamped(long configured, long expected) =>
+        new ScriptingOptions { MaxAllocBytesPerRun = configured }.Sanitized(out _)
+            .MaxAllocBytesPerRun.ShouldBe(expected);
+
+    [Fact]
+    public void An_out_of_range_value_is_reported_as_an_adjustment_naming_the_key()
+    {
+        _ = new ScriptingOptions { StuckOrphanQuarantineFactor = 1e300 }.Sanitized(out var adjustments);
+
+        adjustments.ShouldHaveSingleItem().ShouldContain("Scripting:StuckOrphanQuarantineFactor");
+    }
+
+    [Fact]
+    public void Sanitized_returns_a_copy_and_leaves_the_source_untouched()
+    {
+        var source = new ScriptingOptions { StuckOrphanQuarantineFactor = 1e300 };
+
+        var sanitized = source.Sanitized(out _);
+
+        sanitized.ShouldNotBeSameAs(source);
+        source.StuckOrphanQuarantineFactor.ShouldBe(1e300);
+    }
+
+    /// <summary>
+    ///     The guard takes options from more places than the bind (the Calculation driver's per-driver DTO, tests),
+    ///     so it must not overflow either: an absurd factor reaching the guard directly is clamped, and a refusal
+    ///     behind a young orphan is still a plain refusal rather than an <see cref="OverflowException"/>.
+    /// </summary>
+    [Fact]
+    public async Task An_absurd_factor_reaching_the_guard_directly_does_not_overflow_on_a_refusal()
+    {
+        var guard = new ScriptRunGuard(new ScriptingOptions { StuckOrphanQuarantineFactor = double.MaxValue });
+        const string spin =
+            """
+            var end = System.DateTime.UtcNow.AddSeconds(3);
+            while (System.DateTime.UtcNow < end) { }
+            return 1;
+            """;
+        using var eval = ScriptEvaluator<FakeScriptContext, int>.Compile(spin);
+        var ct = TestContext.Current.CancellationToken;
+        var timeout = TimeSpan.FromMilliseconds(60);
+
+        await Should.ThrowAsync<ScriptTimeoutException>(
+            () => guard.RunAsync("k", spin, eval, new FakeScriptContext(), timeout, ct));
+        // The refusal path evaluates timeout * factor. Unclamped, TimeSpan * double.MaxValue throws OverflowException.
+        await Should.ThrowAsync<ScriptRunRefusedException>(
+            () => guard.RunAsync("k", spin, eval, new FakeScriptContext(), timeout, ct));
+        // A huge timeout times the clamped maximum factor must not overflow either.
+        await Should.ThrowAsync<ScriptRunRefusedException>(
+            () => guard.RunAsync("k", spin, eval, new FakeScriptContext(), TimeSpan.FromDays(365 * 1000), ct));
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.IntegrationTests/AbCipRefusalLiveTests.cs

@@ -0,0 +1,172 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.IntegrationTests;
+
+/// <summary>
+///     Review 07 (Gitea #737) against a running <c>ab_server</c> on both transports: a controller refusal is a per-tag
+///     config error, not a comms fault (H10 probe, H1/L4 data path), and a whole-array write round-trips (U10).
+///     Skipped when the fixture is unreachable (<see cref="AbServerTheoryAttribute"/>); the array write additionally
+///     needs <c>AB_SERVER_ALLOW_WRITES=1</c> (<see cref="IntegrationWriteGate"/>).
+/// </summary>
+[Trait("Category", "Integration")]
+[Trait("Requires", "AbServer")]
+public sealed class AbCipRefusalLiveTests
+{
+    private static string Device(AbServerFixture fixture) => $"ab://{fixture.Host}:{fixture.Port}/1,0";
+
+    private static async Task WaitUntilAsync(Func<bool> condition, TimeSpan timeout, string what)
+    {
+        var deadline = DateTime.UtcNow + timeout;
+        while (!condition())
+        {
+            if (DateTime.UtcNow > deadline) throw new TimeoutException(what);
+            await Task.Delay(50);
+        }
+    }
+
+    /// <summary>
+    ///     H10: a typo'd probe tag on a reachable controller leaves the host Running, and a COLD data tag created
+    ///     after many probe ticks still connects (the backoff window was never opened by the probe's NotFound).
+    /// </summary>
+    /// <param name="transport">The transport stack the driver runs on.</param>
+    [AbServerTheory]
+    [MemberData(nameof(TransportMatrix.Transports), MemberType = typeof(TransportMatrix))]
+    public async Task A_typod_probe_tag_does_not_black_out_a_reachable_device(AbCipTransport transport)
+    {
+        var device = Device(new AbServerFixture());
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(device, AbCipPlcFamily.ControlLogix)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Counter", device, "TestDINT", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions
+            {
+                Enabled = true,
+                ProbeTagPath = "NoSuchProbeTag_xyz",
+                Interval = TimeSpan.FromMilliseconds(200),
+                Timeout = TimeSpan.FromSeconds(2),
+            },
+            Timeout = TimeSpan.FromSeconds(5),
+            Transport = transport,
+        }, $"drv-h10-{transport}");
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            await WaitUntilAsync(() => drv.GetHostStatuses().Single().State == HostState.Running,
+                TimeSpan.FromSeconds(10), "the host never went Running on a refused probe tag");
+            await Task.Delay(1_000); // several probe ticks — each one used to record a Forward-Open failure
+
+            var r = await drv.ReadAsync(["Counter"], CancellationToken.None); // a cold create
+
+            r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+            drv.GetHostStatuses().Single().State.ShouldBe(HostState.Running);
+        }
+        finally
+        {
+            await drv.DisposeAsync();
+        }
+    }
+
+    /// <summary>
+    ///     H1/L4: a tag the controller does not have reads BadNodeIdUnknown every poll while a sibling reads Good, and
+    ///     the driver stays Healthy with LastSuccessfulRead advancing (it used to latch Degraded and, with no other
+    ///     tag, trip the comms-loss watchdog against a reachable controller).
+    /// </summary>
+    /// <param name="transport">The transport stack the driver runs on.</param>
+    [AbServerTheory]
+    [MemberData(nameof(TransportMatrix.Transports), MemberType = typeof(TransportMatrix))]
+    public async Task A_missing_tag_is_per_tag_Bad_and_never_degrades_the_driver(AbCipTransport transport)
+    {
+        var device = Device(new AbServerFixture());
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(device, AbCipPlcFamily.ControlLogix)],
+            RawTags = AbCipRawTags.From(
+                new AbCipTagDefinition("Typo", device, "NoSuchTag_xyz", AbCipDataType.DInt),
+                new AbCipTagDefinition("Counter", device, "TestDINT", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            Timeout = TimeSpan.FromSeconds(5),
+            Transport = transport,
+        }, $"drv-h1-{transport}");
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            DateTime? last = null;
+            for (var i = 0; i < 3; i++)
+            {
+                var r = await drv.ReadAsync(["Typo", "Counter"], CancellationToken.None);
+                r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNodeIdUnknown);
+                r[1].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+                var health = drv.GetHealth();
+                health.State.ShouldBe(DriverState.Healthy);
+                if (last is { } previous) health.LastSuccessfulRead!.Value.ShouldBeGreaterThanOrEqualTo(previous);
+                last = health.LastSuccessfulRead;
+            }
+
+            // Only the refused tag: still Healthy, still answering (the watchdog conjunction can never trip).
+            var before = drv.GetHealth().LastSuccessfulRead!.Value;
+            await Task.Delay(20);
+            (await drv.ReadAsync(["Typo"], CancellationToken.None))[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNodeIdUnknown);
+            drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+            drv.GetHealth().LastSuccessfulRead!.Value.ShouldBeGreaterThan(before);
+        }
+        finally
+        {
+            await drv.DisposeAsync();
+        }
+    }
+
+    /// <summary>
+    ///     U10: a whole <c>int[]</c> written to the 16-element <c>TestArray</c> (DINT[16]) reads back element for element
+    ///     through a fresh driver; a wrong-length write is refused BadOutOfRange and changes nothing.
+    /// </summary>
+    /// <param name="transport">The transport stack the driver runs on.</param>
+    [AbServerTheory]
+    [MemberData(nameof(TransportMatrix.Transports), MemberType = typeof(TransportMatrix))]
+    public async Task A_whole_array_write_round_trips(AbCipTransport transport)
+    {
+        IntegrationWriteGate.SkipUnlessWritesAllowed();
+        var device = Device(new AbServerFixture());
+        var seed = Random.Shared.Next(1_000, 1_000_000);
+        var values = Enumerable.Range(0, 16).Select(i => seed + i * 7).ToArray();
+
+        AbCipDriver Make(string id) => new(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(device, AbCipPlcFamily.ControlLogix)],
+            RawTags = AbCipRawTags.From(
+                new AbCipTagDefinition("Arr", device, "TestArray", AbCipDataType.DInt, ElementCount: 16, IsArray: true)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            Timeout = TimeSpan.FromSeconds(5),
+            Transport = transport,
+        }, $"drv-u10-{id}-{transport}");
+
+        var writer = Make("writer");
+        await writer.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            var w = await writer.WriteAsync([new WriteRequest("Arr", values)], CancellationToken.None);
+            w[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+            var bad = await writer.WriteAsync([new WriteRequest("Arr", values.Take(15).ToArray())], CancellationToken.None);
+            bad[0].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+            writer.GetHealth().State.ShouldBe(DriverState.Healthy);
+        }
+        finally
+        {
+            await writer.DisposeAsync();
+        }
+
+        var reader = Make("reader");
+        await reader.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            var r = await reader.ReadAsync(["Arr"], CancellationToken.None);
+            r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+            r[0].Value.ShouldBe(values);
+        }
+        finally
+        {
+            await reader.DisposeAsync();
+        }
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipAlmaProjectionTests.cs

@@ -311,8 +311,9 @@ public sealed class AbCipAlmaProjectionTests
         // A non-zero libplctag status ⇒ Bad on every member read; the driver evicts the handle and re-creates it, so
         // every re-created runtime must fail too (a fresh fake would read as "all clear").
         var recreations = 0;
-        r.Factory.Customise = p => { Interlocked.Increment(ref recreations); return new FakeAbCipTag(p) { Status = (int)AbStatus.NotFound }; };
-        foreach (var t in r.Factory.Tags.Values.ToList()) t.Status = (int)AbStatus.NotFound;
+        // A comms status (not a device refusal, which keeps the handle since review 07 H1) so the handle is evicted.
+        r.Factory.Customise = p => { Interlocked.Increment(ref recreations); return new FakeAbCipTag(p) { Status = (int)AbStatus.BadConnection }; };
+        foreach (var t in r.Factory.Tags.Values.ToList()) t.Status = (int)AbStatus.BadConnection;
         await WaitFor(() => Volatile.Read(ref recreations) >= 6);
         await r.StopAsync();
         Snapshot(r).Any(x => x.Kind == AlarmTransitionKind.Clear).ShouldBeFalse();
@@ -334,9 +335,9 @@ public sealed class AbCipAlmaProjectionTests
         {
             if (!p.TagName.EndsWith("HAcked", StringComparison.Ordinal)) return null!;
             Interlocked.Increment(ref bad);
-            return new FakeAbCipTag(p) { Status = (int)AbStatus.NotFound };
+            return new FakeAbCipTag(p) { Status = (int)AbStatus.BadConnection }; // comms ⇒ evicted + re-created
         };
-        r.Factory.Tags[$"{Tag}.HAcked"].Status = (int)AbStatus.NotFound;
+        r.Factory.Tags[$"{Tag}.HAcked"].Status = (int)AbStatus.BadConnection;
         await WaitFor(() => Volatile.Read(ref bad) >= 4);
         // The link recovers and HAcked reads true again.
         r.Factory.Customise = p => new FakeAbCipTag(p) { Value = true };

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipArrayWriteTests.cs

@@ -0,0 +1,179 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport.Tests.Fakes;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Review 07 U10: array writes. Config and the AdminUI allowed them and discovery advertised array nodes
+///     Operate-writable, but both runtimes encoded a SCALAR: an <c>int[]</c> threw InvalidCastException (BadTypeMismatch
+///     AND Degraded), a scalar wrote element 0 natively, and libplctag re-sent the whole last-read buffer with only
+///     element 0 patched. The driver now encodes <c>count</c> elements at the element width, refuses a length mismatch
+///     with <c>BadOutOfRange</c>, and converts element values (ScadaBridge writes <c>int[]</c> to INT arrays).
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipArrayWriteTests
+{
+    private const string Device = "ab://10.0.0.5/1,0";
+
+    private static async Task<(AbCipDriver Driver, FakeAbCipTagFactory Factory)> BuildAsync(
+        AbCipDataType type, int count, int size = -1)
+    {
+        var factory = new FakeAbCipTagFactory { Customise = p => new FakeAbCipTag(p) { Size = size } };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Arr", Device, "Arr", type, ElementCount: count, IsArray: true)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+        }, "abcip-array-write", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        return (drv, factory);
+    }
+
+    // ---- the codec ----
+
+    /// <summary>Each element is converted and laid out little-endian at the element width.</summary>
+    [Fact]
+    public void The_codec_converts_elements_and_lays_them_out_at_the_element_width()
+    {
+        AbCipArrayCodec.Encode(AbCipDataType.Int, [1, -2, 300]).ShouldBe(Convert.FromHexString("0100FEFF2C01"));
+        AbCipArrayCodec.Encode(AbCipDataType.SInt, [(short)-1, 5L]).ShouldBe(Convert.FromHexString("FF05"));
+        AbCipArrayCodec.Encode(AbCipDataType.DInt, [1.0, 2]).ShouldBe(Convert.FromHexString("0100000002000000"));
+        AbCipArrayCodec.Encode(AbCipDataType.Real, [1, 2.5]).ShouldBe(Convert.FromHexString("0000803F00002040"));
+        AbCipArrayCodec.Encode(AbCipDataType.ULInt, [ulong.MaxValue]).ShouldBe(Convert.FromHexString("FFFFFFFFFFFFFFFF"));
+        Should.Throw<OverflowException>(() => AbCipArrayCodec.Encode(AbCipDataType.Int, [40000]));
+        Should.Throw<NotSupportedException>(() => AbCipArrayCodec.Encode(AbCipDataType.Bool, [true]));
+    }
+
+    // ---- the driver ----
+
+    /// <summary>ScadaBridge's wire shape: an <c>int[]</c> to an INT[3] encodes 3 two-byte elements and writes once, Good.</summary>
+    [Fact]
+    public async Task An_int_array_to_an_INT_array_encodes_every_element()
+    {
+        var (drv, factory) = await BuildAsync(AbCipDataType.Int, 3, size: 6);
+
+        var r = await drv.WriteAsync([new WriteRequest("Arr", new[] { 1, 2, 3 })], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        var tag = factory.Tags["Arr"];
+        tag.EncodedArrayType.ShouldBe(AbCipDataType.Int);
+        tag.EncodedArrayCount.ShouldBe(3);
+        tag.EncodedArrayBytes.ShouldBe(Convert.FromHexString("010002000300"));
+        tag.WriteCount.ShouldBe(1);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>A list of a different length is BadOutOfRange; nothing is created, encoded or written.</summary>
+    /// <param name="length">The client's element count.</param>
+    [Theory]
+    [InlineData(2)]
+    [InlineData(4)]
+    public async Task A_length_mismatch_is_BadOutOfRange_and_nothing_is_written(int length)
+    {
+        var (drv, factory) = await BuildAsync(AbCipDataType.DInt, 3);
+
+        var r = await drv.WriteAsync([new WriteRequest("Arr", Enumerable.Range(0, length).ToArray())], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+        factory.Tags.ShouldNotContainKey("Arr"); // refused before any runtime exists
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>A scalar (or a string) to an array tag is BadTypeMismatch — it used to write element 0 only.</summary>
+    [Fact]
+    public async Task A_scalar_to_an_array_tag_is_BadTypeMismatch()
+    {
+        var (drv, factory) = await BuildAsync(AbCipDataType.DInt, 3);
+
+        var r = await drv.WriteAsync(
+            [new WriteRequest("Arr", 5), new WriteRequest("Arr", "1,2,3")], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadTypeMismatch);
+        r[1].StatusCode.ShouldBe(AbCipStatusMapper.BadTypeMismatch);
+        factory.Tags.ShouldNotContainKey("Arr");
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>An element that does not fit is BadOutOfRange with no health write and nothing sent.</summary>
+    [Fact]
+    public async Task An_element_overflow_is_BadOutOfRange_and_nothing_is_sent()
+    {
+        var (drv, factory) = await BuildAsync(AbCipDataType.Int, 2, size: 4);
+
+        var r = await drv.WriteAsync([new WriteRequest("Arr", new[] { 1, 70000 })], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+        factory.Tags["Arr"].WriteCount.ShouldBe(0);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>BOOL / STRING arrays are refused BadNotSupported before any runtime exists.</summary>
+    /// <param name="type">The array element type.</param>
+    [Theory]
+    [InlineData(AbCipDataType.Bool)]
+    [InlineData(AbCipDataType.String)]
+    public async Task Unencodable_element_types_are_BadNotSupported(AbCipDataType type)
+    {
+        var (drv, factory) = await BuildAsync(type, 2);
+
+        var r = await drv.WriteAsync([new WriteRequest("Arr", new object[] { true, false })], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotSupported);
+        factory.Tags.ShouldNotContainKey("Arr");
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy); // M5: a refusal writes no health
+    }
+
+    /// <summary>The controller holding a different shape than authored (size ≠ width × count) is BadTypeMismatch, nothing sent.</summary>
+    [Fact]
+    public async Task A_controller_shape_mismatch_is_BadTypeMismatch_and_nothing_is_sent()
+    {
+        var (drv, factory) = await BuildAsync(AbCipDataType.Int, 3, size: 12); // DINT[3] on the controller
+
+        var r = await drv.WriteAsync([new WriteRequest("Arr", new[] { 1, 2, 3 })], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadTypeMismatch);
+        factory.Tags["Arr"].WriteCount.ShouldBe(0);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>
+    ///     End to end over the REAL native runtime: an <c>int[]</c> to an INT[3] goes out as ONE Write Tag of type
+    ///     0xC3 with element count 3 and the three 2-byte elements.
+    /// </summary>
+    [Fact]
+    public async Task A_native_array_write_is_one_write_of_count_elements()
+    {
+        await using var target = new FakeCipTarget();
+        target.ConnectedHandler = req => Task.FromResult<byte[]?>(req[0] switch
+        {
+            0x52 => [0xD2, 0x00, 0x00, 0x00, 0xC3, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00], // INT[3] = 0,0,0
+            0x4D => [0xCD, 0x00, 0x00, 0x00],
+            _ => target.ServiceReply(req),
+        });
+        var host = $"ab://127.0.0.1:{target.Port}/1,0";
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(host, AbCipPlcFamily.ControlLogix)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Arr", host, "Arr", AbCipDataType.Int, ElementCount: 3, IsArray: true)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            Timeout = TimeSpan.FromSeconds(5),
+        }, "abcip-array-native", new NativeAbCipTagFactory(new DeviceSessionPool()));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            var r = await drv.WriteAsync([new WriteRequest("Arr", new[] { 1, -2, 300 })], CancellationToken.None);
+
+            r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+            var write = target.ConnectedCips.Single(c => c[0] == 0x4D);
+            var typeAt = 2 + write[1] * 2;
+            write.AsSpan(typeAt).ToArray().ShouldBe(Convert.FromHexString("C300" + "0300" + "0100FEFF2C01"));
+        }
+        finally
+        {
+            await drv.DisposeAsync();
+        }
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipBitInWordWidthTests.cs

@@ -0,0 +1,161 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport.Tests.Fakes;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Review 07 H17: Logix allows a <c>.n</c> bit selector on SINT / INT / DINT (and LINT) hosts, and reads of
+///     <c>Flags.3</c> on an INT worked — but the bit write's read-modify-write hard-coded a DINT parent, so its
+///     <c>DecodeValue(DInt)</c> on a 2-byte buffer threw OutOfBounds and the write was reported as a comms fault
+///     (BadCommunicationError + Degraded). The parent width now comes from the parent read's size.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipBitInWordWidthTests
+{
+    private const string Device = "ab://10.0.0.5/1,0";
+
+    private static async Task<(AbCipDriver Driver, FakeAbCipTagFactory Factory)> BuildAsync(int parentSize, object parentValue, string bitPath)
+    {
+        var factory = new FakeAbCipTagFactory
+        {
+            Customise = p => new FakeAbCipTag(p) { Value = parentValue, Size = parentSize },
+        };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Bit", Device, bitPath, AbCipDataType.Bool)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+        }, "abcip-bit", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        return (drv, factory);
+    }
+
+    /// <summary>SINT / INT / DINT hosts: the bit is set, every other bit is preserved, nothing degrades.</summary>
+    /// <param name="size">The parent's wire width.</param>
+    /// <param name="before">The parent's value before the write.</param>
+    /// <param name="bit">The bit written true.</param>
+    /// <param name="after">The value the parent must be written back with.</param>
+    [Theory]
+    [InlineData(1, 0x81, 3, 0x89)]       // SINT 0x81 (-127): bit 7 set stays set, written as its unsigned byte
+    [InlineData(1, 0x00, 7, 0x80)]       // SINT top bit: 128 would overflow an sbyte encode
+    [InlineData(2, 0x8001, 3, 0x8009)]   // INT top bit preserved
+    [InlineData(2, 0x0000, 15, 0x8000)]  // INT top bit: 32768 would overflow a short encode
+    [InlineData(4, 0x00000001, 3, 0x00000009)]
+    public async Task A_bit_write_uses_the_parent_width(int size, int before, int bit, int after)
+    {
+        var (drv, factory) = await BuildAsync(size, before, $"Motor.Flags.{bit}");
+
+        var r = await drv.WriteAsync([new WriteRequest("Bit", true)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        Convert.ToInt64(factory.Tags["Motor.Flags"].Value).ShouldBe(after);
+        factory.Tags["Motor.Flags"].WriteCount.ShouldBe(1);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>A bit past the host's width is refused before anything is written, with no health write.</summary>
+    [Fact]
+    public async Task A_bit_past_the_host_width_is_BadOutOfRange_and_nothing_is_written()
+    {
+        var (drv, factory) = await BuildAsync(2, 0, "Motor.Flags.20");
+
+        var r = await drv.WriteAsync([new WriteRequest("Bit", true)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+        factory.Tags["Motor.Flags"].WriteCount.ShouldBe(0);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>A host whose width is no integer's is a type mismatch, refused before writing, no health write.</summary>
+    [Fact]
+    public async Task A_host_of_no_integer_width_is_BadTypeMismatch()
+    {
+        var (drv, factory) = await BuildAsync(3, 0, "Motor.Flags.1");
+
+        var r = await drv.WriteAsync([new WriteRequest("Bit", true)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadTypeMismatch);
+        factory.Tags["Motor.Flags"].WriteCount.ShouldBe(0);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>
+    ///     End to end over the REAL native runtime against a fake CIP target whose <c>Flags</c> is an INT: the write
+    ///     goes out as ONE INT (type 0xC3, two data bytes) with the bit set and the other bits preserved.
+    /// </summary>
+    [Fact]
+    public async Task A_native_bit_write_on_an_INT_host_writes_an_INT()
+    {
+        await using var target = new FakeCipTarget();
+        // Read Tag Fragmented (0x52) → INT (0x00C3) 0x8001; Write Tag (0x4D) → success.
+        target.ConnectedHandler = req => Task.FromResult<byte[]?>(req[0] switch
+        {
+            0x52 => [0xD2, 0x00, 0x00, 0x00, 0xC3, 0x00, 0x01, 0x80],
+            0x4D => [0xCD, 0x00, 0x00, 0x00],
+            _ => target.ServiceReply(req),
+        });
+        var host = $"ab://127.0.0.1:{target.Port}/1,0";
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(host, AbCipPlcFamily.ControlLogix)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Bit", host, "Flags.3", AbCipDataType.Bool)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            Timeout = TimeSpan.FromSeconds(5),
+        }, "abcip-bit-native", new NativeAbCipTagFactory(new DeviceSessionPool()));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            var r = await drv.WriteAsync([new WriteRequest("Bit", true)], CancellationToken.None);
+
+            r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+            var write = target.ConnectedCips.Single(c => c[0] == 0x4D);
+            var typeAt = 2 + write[1] * 2;
+            write[typeAt].ShouldBe((byte)0xC3);                 // INT, not DINT
+            write.Length.ShouldBe(typeAt + 2 + 2 + 2);          // type, element count, ONE 2-byte element
+            write[^2].ShouldBe((byte)0x09);                     // 0x8001 | bit 3
+            write[^1].ShouldBe((byte)0x80);
+            drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+        }
+        finally
+        {
+            await drv.DisposeAsync();
+        }
+    }
+
+    /// <summary>
+    ///     M1: a parent read that fails with a COMMS status evicts the parent runtime and degrades (LastSuccessfulRead
+    ///     preserved); one the controller REFUSES is per-tag Bad, keeps the runtime and writes no health.
+    /// </summary>
+    [Fact]
+    public async Task A_failed_parent_read_follows_the_health_contract()
+    {
+        var (drv, factory) = await BuildAsync(4, 0, "Motor.Flags.3");
+        factory.Customise = p => new FakeAbCipTag(p) { Status = (int)AbStatus.BadConnection };
+        var before = drv.GetHealth();
+
+        var r = await drv.WriteAsync([new WriteRequest("Bit", true)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadCommunicationError);
+        factory.Tags["Motor.Flags"].Disposed.ShouldBeTrue();
+        factory.Tags["Motor.Flags"].WriteCount.ShouldBe(0);
+        drv.GetHealth().State.ShouldBe(DriverState.Degraded);
+        drv.GetHealth().LastSuccessfulRead.ShouldBe(before.LastSuccessfulRead);
+    }
+
+    /// <summary>M1 positive control: a refused parent read keeps the runtime, no health write.</summary>
+    [Fact]
+    public async Task A_refused_parent_read_keeps_the_runtime_and_health()
+    {
+        var (drv, factory) = await BuildAsync(4, 0, "Motor.Flags.3");
+        factory.Customise = p => new FakeAbCipTag(p) { Status = (int)AbStatus.NotAllowed };
+
+        var r = await drv.WriteAsync([new WriteRequest("Bit", true)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotWritable);
+        factory.Tags["Motor.Flags"].Disposed.ShouldBeFalse();
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipDeviceRefusalTests.cs

@@ -0,0 +1,298 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport.Tests.Fakes;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Review 07 H1 / L4: a status that proves the controller ANSWERED (NotFound, OutOfBounds, NotAllowed,
+///     TooLarge, TooSmall, TypeMismatch, Unsupported) is a deterministic per-tag refusal, not a comms fault. It
+///     keeps the tag runtime (no evict → no lease / Forward-Open churn), writes no Degraded, opens no device-wide
+///     backoff window for siblings, warns once, and — since the device answered — advances LastSuccessfulRead.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipDeviceRefusalTests
+{
+    private const string Device = "ab://10.0.0.5/1,0";
+
+    /// <summary>A fake whose reads/writes throw a chosen transport status (init succeeds unless told otherwise).</summary>
+    private sealed class RefusingTag(AbCipTagCreateParams p) : FakeAbCipTag(p)
+    {
+        public AbStatus? ReadThrows { get; set; }
+        public AbStatus? WriteThrows { get; set; }
+
+        public override async Task ReadAsync(CancellationToken cancellationToken)
+        {
+            await base.ReadAsync(cancellationToken);
+            if (ReadThrows is { } s) throw new AbTransportException(s, s.ToString());
+        }
+
+        public override async Task WriteAsync(CancellationToken cancellationToken)
+        {
+            await base.WriteAsync(cancellationToken);
+            if (WriteThrows is { } s) throw new AbTransportException(s, s.ToString());
+        }
+    }
+
+    private static (AbCipDriver Driver, FakeAbCipTagFactory Factory) Build(
+        Func<AbCipTagCreateParams, FakeAbCipTag> make)
+    {
+        var factory = new FakeAbCipTagFactory { Customise = make };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(
+                new AbCipTagDefinition("Bad", Device, "Arr[999]", AbCipDataType.DInt),
+                new AbCipTagDefinition("Good", Device, "Real1", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            MaxConcurrentReads = 1,
+        }, "abcip-refusal", factory);
+        return (drv, factory);
+    }
+
+    /// <summary>The classification itself: every answered refusal is reachable; transport failures are not.</summary>
+    /// <param name="status">The transport status.</param>
+    /// <param name="refusal">Whether it is a device refusal.</param>
+    [Theory]
+    [InlineData(AbStatus.NotFound, true)]
+    [InlineData(AbStatus.OutOfBounds, true)]
+    [InlineData(AbStatus.NotAllowed, true)]
+    [InlineData(AbStatus.TooLarge, true)]
+    [InlineData(AbStatus.TooSmall, true)]
+    [InlineData(AbStatus.TypeMismatch, true)]
+    [InlineData(AbStatus.Unsupported, true)]
+    [InlineData(AbStatus.Ok, false)]
+    [InlineData(AbStatus.Timeout, false)]
+    [InlineData(AbStatus.BadConnection, false)]
+    [InlineData(AbStatus.BadReply, false)]
+    [InlineData(AbStatus.Partial, false)]
+    [InlineData(AbStatus.Abort, false)]
+    public void Answered_statuses_are_device_refusals(AbStatus status, bool refusal)
+    {
+        AbCipRefusals.IsDeviceRefusal(status).ShouldBe(refusal);
+        AbCipDriverProbe.IsReachableStatus(status).ShouldBe(refusal || status == AbStatus.Ok);
+    }
+
+    /// <summary>A read the controller refuses keeps its runtime, stays Healthy, and advances LastSuccessfulRead.</summary>
+    /// <param name="status">The refusal the controller answers with.</param>
+    /// <param name="expected">The OPC UA status the tag reads.</param>
+    [Theory]
+    [InlineData(AbStatus.NotFound, AbCipStatusMapper.BadNodeIdUnknown)]
+    [InlineData(AbStatus.OutOfBounds, AbCipStatusMapper.BadOutOfRange)]
+    [InlineData(AbStatus.TooLarge, AbCipStatusMapper.BadOutOfRange)]
+    [InlineData(AbStatus.TooSmall, AbCipStatusMapper.BadOutOfRange)]
+    [InlineData(AbStatus.NotAllowed, AbCipStatusMapper.BadNotReadable)]
+    public async Task A_refused_read_keeps_the_runtime_and_writes_no_Degraded(AbStatus status, uint expected)
+    {
+        var (drv, factory) = Build(p => new RefusingTag(p) { ReadThrows = p.TagName == "Arr[999]" ? status : null, Value = 1 });
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        var before = drv.GetHealth().LastSuccessfulRead!.Value;
+        await Task.Delay(20);
+
+        for (var i = 0; i < 3; i++)
+        {
+            var r = await drv.ReadAsync(["Bad"], CancellationToken.None);
+            r[0].StatusCode.ShouldBe(expected);
+        }
+
+        var bad = factory.Tags["Arr[999]"];
+        bad.InitializeCount.ShouldBe(1); // never evicted → never re-created
+        bad.Disposed.ShouldBeFalse();
+        bad.ReadCount.ShouldBe(3);
+        var health = drv.GetHealth();
+        health.State.ShouldBe(DriverState.Healthy);
+        health.LastSuccessfulRead!.Value.ShouldBeGreaterThan(before); // the device answered
+    }
+
+    /// <summary>Positive control: a genuine transport failure on the same tag still evicts and degrades.</summary>
+    [Fact]
+    public async Task A_transport_failure_still_evicts_and_degrades()
+    {
+        var (drv, factory) = Build(p => new RefusingTag(p) { ReadThrows = p.TagName == "Arr[999]" ? AbStatus.Timeout : null });
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        var r = await drv.ReadAsync(["Bad"], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadTimeout);
+        factory.Tags["Arr[999]"].Disposed.ShouldBeTrue();
+        drv.GetHealth().State.ShouldBe(DriverState.Degraded);
+    }
+
+    /// <summary>A refusal returned as a status word (no throw) is classified the same way.</summary>
+    [Fact]
+    public async Task A_refusal_status_word_keeps_the_runtime()
+    {
+        var (drv, factory) = Build(p => new FakeAbCipTag(p) { Status = p.TagName == "Arr[999]" ? (int)AbStatus.OutOfBounds : 0, Value = 1 });
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        await drv.ReadAsync(["Bad"], CancellationToken.None);
+        var r = await drv.ReadAsync(["Bad", "Good"], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+        r[1].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        factory.Tags["Arr[999]"].InitializeCount.ShouldBe(1);
+        factory.Tags["Arr[999]"].Disposed.ShouldBeFalse();
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>
+    ///     A cold create the controller refuses with OutOfBounds proves the session works: it must not open the
+    ///     device-wide backoff window, so a sibling cold tag still connects and reads Good.
+    /// </summary>
+    [Fact]
+    public async Task A_refused_create_does_not_throttle_sibling_cold_creates()
+    {
+        var (drv, _) = Build(p => p.TagName == "Arr[999]"
+            ? new FakeAbCipTag(p) { ThrowOnInitialize = true, Exception = new AbTransportException(AbStatus.OutOfBounds, "OutOfBounds") }
+            : new FakeAbCipTag(p) { Value = 42 });
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        var r = await drv.ReadAsync(["Bad", "Good"], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+        r[1].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        r[1].Value.ShouldBe(42);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>A write the controller refuses keeps the runtime and writes no health.</summary>
+    [Fact]
+    public async Task A_refused_write_keeps_the_runtime_and_writes_no_Degraded()
+    {
+        var factory = new FakeAbCipTagFactory { Customise = p => new RefusingTag(p) { WriteThrows = AbStatus.NotAllowed } };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Ro", Device, "ReadOnlyTag", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+        }, "abcip-refusal", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        var w1 = await drv.WriteAsync([new WriteRequest("Ro", 5)], CancellationToken.None);
+        var w2 = await drv.WriteAsync([new WriteRequest("Ro", 6)], CancellationToken.None);
+
+        w1[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotWritable);
+        w2[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotWritable);
+        factory.Tags["ReadOnlyTag"].InitializeCount.ShouldBe(1);
+        factory.Tags["ReadOnlyTag"].Disposed.ShouldBeFalse();
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>
+    ///     L4 end to end over the REAL native runtime + a private pool: a renamed tag (CIP 0x04) that is the device's
+    ///     ONLY lease holder used to be evicted every poll — the last release, so Forward Close, then a throttled Forward
+    ///     Open on the next poll. One runtime and one Forward Open across N polls and no Forward Close proves no churn.
+    /// </summary>
+    [Fact]
+    public async Task A_native_NotFound_tag_causes_no_runtime_or_connection_churn()
+    {
+        await using var target = new FakeCipTarget();
+        target.ConnectedHandler = req => Task.FromResult<byte[]?>(
+            req[0] == 0x52 ? [0xD2, 0x00, 0x04, 0x00] : target.ServiceReply(req));
+        var host = $"ab://127.0.0.1:{target.Port}/1,0";
+        var creates = 0;
+        var native = new NativeAbCipTagFactory(new DeviceSessionPool());
+        var factory = new CountingFactory(p => { Interlocked.Increment(ref creates); return native.Create(p); });
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(host, AbCipPlcFamily.ControlLogix)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Renamed", host, "Renamed", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            Timeout = TimeSpan.FromSeconds(5),
+        }, "abcip-l4-native", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        try
+        {
+            for (var n = 0; n < 5; n++)
+                (await drv.ReadAsync(["Renamed"], CancellationToken.None))[0].StatusCode
+                    .ShouldBe(AbCipStatusMapper.BadNodeIdUnknown);
+
+            Volatile.Read(ref creates).ShouldBe(1);
+            target.ForwardOpens.ShouldBe(1);
+            target.ForwardCloses.ShouldBe(0);
+            target.ConnectedRequests.ShouldBeGreaterThanOrEqualTo(5); // positive control: every poll hit the wire
+            drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+        }
+        finally
+        {
+            await drv.DisposeAsync();
+        }
+    }
+
+    private sealed class CountingFactory(Func<AbCipTagCreateParams, IAbCipTagRuntime> create) : IAbCipTagFactory
+    {
+        public IAbCipTagRuntime Create(AbCipTagCreateParams createParams) => create(createParams);
+    }
+
+    /// <summary>
+    ///     #737 I2: a LOCAL refusal (libplctag ErrorNotImplemented — no I/O) is per-tag BadNotSupported, keeps health
+    ///     state, and does NOT advance LastSuccessfulRead (nothing proves the device answered). Positive control: the
+    ///     device's own Unsupported refusal does advance it (M2).
+    /// </summary>
+    [Fact]
+    public async Task A_local_refusal_does_not_advance_LastSuccessfulRead_but_a_device_one_does()
+    {
+        var (drv, _) = Build(p => p.TagName == "Arr[999]"
+            ? new LocalRefusingTag(p)
+            : new RefusingTag(p) { ReadThrows = AbStatus.Unsupported });
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        var before = drv.GetHealth().LastSuccessfulRead!.Value;
+        await Task.Delay(20);
+
+        (await drv.ReadAsync(["Bad"], CancellationToken.None))[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotSupported);
+        drv.GetHealth().LastSuccessfulRead.ShouldBe(before);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+
+        (await drv.ReadAsync(["Good"], CancellationToken.None))[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotSupported);
+        drv.GetHealth().LastSuccessfulRead!.Value.ShouldBeGreaterThan(before);
+    }
+
+    /// <summary>#737 I2: a local refusal at create opens no backoff — a sibling cold create still connects.</summary>
+    [Fact]
+    public async Task A_local_refusal_at_create_does_not_throttle_or_count_as_answered()
+    {
+        var (drv, _) = Build(p => p.TagName == "Arr[999]"
+            ? new FakeAbCipTag(p) { ThrowOnInitialize = true, Exception = new AbCipLocalRefusalException("ErrorNotImplemented") }
+            : new FakeAbCipTag(p) { Value = 3 });
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        var r = await drv.ReadAsync(["Bad", "Good"], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNotSupported);
+        r[1].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    /// <summary>M2: a write the device refuses proves it answered — LastSuccessfulRead advances.</summary>
+    [Fact]
+    public async Task A_refused_write_advances_LastSuccessfulRead()
+    {
+        var factory = new FakeAbCipTagFactory { Customise = p => new RefusingTag(p) { WriteThrows = AbStatus.NotAllowed } };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Ro", Device, "ReadOnlyTag", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+        }, "abcip-refusal", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        var before = drv.GetHealth().LastSuccessfulRead!.Value;
+        await Task.Delay(20);
+
+        (await drv.WriteAsync([new WriteRequest("Ro", 5)], CancellationToken.None))[0].StatusCode
+            .ShouldBe(AbCipStatusMapper.BadNotWritable);
+
+        drv.GetHealth().LastSuccessfulRead!.Value.ShouldBeGreaterThan(before);
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+
+    private sealed class LocalRefusingTag(AbCipTagCreateParams p) : FakeAbCipTag(p)
+    {
+        public override async Task ReadAsync(CancellationToken cancellationToken)
+        {
+            await base.ReadAsync(cancellationToken);
+            throw new AbCipLocalRefusalException("ErrorNotImplemented");
+        }
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipLifecycleConcurrencyTests.cs

@@ -0,0 +1,239 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Lifecycle overlap the runtime can produce (review 07 H13/F10, Lane A): the final stop runs
+///     <c>ShutdownAsync</c> then <c>DisposeAsync</c> under one bounded budget, so a slow shutdown is abandoned and
+///     the dispose runs CONCURRENTLY with it; an abandoned <c>InitializeAsync</c> can overlap the next attempt's
+///     shutdown. Device tables are build-and-swapped, so each teardown owns a disjoint set of device states.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipLifecycleConcurrencyTests
+{
+    private const string Host = "ab://10.0.0.5/1,0";
+
+    /// <summary>A fake whose reads ignore cancellation for a while, so a teardown waits on the probe loop.</summary>
+    private sealed class SlowTag(AbCipTagCreateParams p, TimeSpan readDelay) : FakeAbCipTag(p)
+    {
+        private int _disposeCount;
+
+        public int DisposeCount => Volatile.Read(ref _disposeCount);
+
+        public override async Task ReadAsync(CancellationToken cancellationToken)
+        {
+            await base.ReadAsync(cancellationToken);
+            if (readDelay > TimeSpan.Zero) await Task.Delay(readDelay, CancellationToken.None);
+        }
+
+        public override void Dispose()
+        {
+            Interlocked.Increment(ref _disposeCount);
+            base.Dispose();
+        }
+    }
+
+    private static (AbCipDriver Driver, FakeAbCipTagFactory Factory) Build(TimeSpan probeReadDelay)
+    {
+        var factory = new FakeAbCipTagFactory
+        {
+            Customise = p => new SlowTag(p, p.TagName == "ProbeTag" ? probeReadDelay : TimeSpan.Zero) { Value = 1 },
+        };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Host)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Speed", Host, "Motor1.Speed", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions
+            {
+                Enabled = true,
+                ProbeTagPath = "ProbeTag",
+                Interval = TimeSpan.FromMilliseconds(10),
+                Timeout = TimeSpan.FromMilliseconds(200),
+            },
+        }, "abcip-lifecycle", factory);
+        return (drv, factory);
+    }
+
+    private static async Task WaitUntilAsync(Func<bool> condition, string what, int seconds = 5)
+    {
+        var deadline = DateTime.UtcNow.AddSeconds(seconds);
+        while (!condition())
+        {
+            if (DateTime.UtcNow > deadline) throw new TimeoutException(what);
+            await Task.Delay(5);
+        }
+    }
+
+    /// <summary>
+    ///     Shutdown and Dispose racing each other tear the device down exactly once: the data runtime is
+    ///     disposed once (not once per teardown), neither call throws, and the probe loop has exited.
+    /// </summary>
+    [Fact]
+    public async Task Dispose_concurrent_with_a_running_Shutdown_tears_each_state_down_exactly_once()
+    {
+        var (drv, factory) = Build(probeReadDelay: TimeSpan.FromMilliseconds(300));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        (await drv.ReadAsync(["Speed"], CancellationToken.None))[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        await WaitUntilAsync(() => factory.Tags.ContainsKey("ProbeTag"), "probe never started");
+        var state = drv.GetDeviceState(Host)!;
+        var probeLoop = state.ProbeTask!;
+        var data = (SlowTag)factory.Tags["Motor1.Speed"];
+
+        var shutdown = drv.ShutdownAsync(CancellationToken.None);
+        var dispose = drv.DisposeAsync().AsTask();
+        await Task.WhenAll(shutdown, dispose);
+
+        data.DisposeCount.ShouldBe(1);
+        probeLoop.IsCompleted.ShouldBeTrue();
+        drv.GetDeviceState(Host).ShouldBeNull();
+    }
+
+    /// <summary>
+    ///     An InitializeAsync that a ShutdownAsync overtakes publishes nothing: no device state and no probe
+    ///     loop survive the shutdown. Positive control: an init that nothing overtakes still publishes.
+    /// </summary>
+    [Fact]
+    public async Task An_init_overtaken_by_Shutdown_publishes_no_device_and_starts_no_probe()
+    {
+        var (drv, factory) = Build(probeReadDelay: TimeSpan.FromMilliseconds(300));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        await WaitUntilAsync(() => factory.Tags.ContainsKey("ProbeTag"), "probe never started");
+        var firstProbe = factory.Tags["ProbeTag"];
+
+        // The second init's own release waits on the slow first probe loop; the shutdown overtakes it there.
+        var init = drv.InitializeAsync("{}", CancellationToken.None);
+        await Task.Delay(50);
+        await drv.ShutdownAsync(CancellationToken.None);
+        var superseded = await Should.ThrowAsync<AbCipInitializationSupersededException>(() => init);
+        superseded.ShouldNotBeAssignableTo<OperationCanceledException>(); // M4: distinct from a caller cancellation
+
+        drv.GetDeviceState(Host).ShouldBeNull();
+        drv.GetHostStatuses().ShouldBeEmpty();
+        await Task.Delay(100);
+        factory.Tags["ProbeTag"].ShouldBeSameAs(firstProbe); // no second probe loop ever created a runtime
+        drv.GetHealth().State.ShouldNotBe(DriverState.Healthy);
+
+        // Positive control: a later init that nothing overtakes publishes the device again.
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        drv.GetDeviceState(Host).ShouldNotBeNull();
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+        await drv.DisposeAsync();
+    }
+
+    /// <summary>
+    ///     Readers of the device/routing tables (the retained poll engine, the actor's host-status pull) never
+    ///     see a table mid-mutation while the driver re-initialises underneath them, never trip over a per-tag lock a
+    ///     retire disposed, and never leak a runtime they cold-created into a device state a retire already tore down.
+    /// </summary>
+    [Fact]
+    public async Task Readers_survive_repeated_reinitialisation_and_leak_no_runtime()
+    {
+        var created = new System.Collections.Concurrent.ConcurrentBag<FakeAbCipTag>();
+        var factory = new FakeAbCipTagFactory
+        {
+            Customise = p =>
+            {
+                var t = new FakeAbCipTag(p) { Value = 1 };
+                created.Add(t);
+                return t;
+            },
+        };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Host)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Speed", Host, "Motor1.Speed", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+        }, "abcip-lifecycle", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        using var stop = new CancellationTokenSource();
+        var started = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
+        var badHealth = 0;
+        var reader = Task.Run(async () =>
+        {
+            var reads = 0;
+            while (!stop.IsCancellationRequested)
+            {
+                _ = drv.GetHostStatuses();
+                _ = drv.ResolveHost("Speed");
+                await drv.ReadAsync(["Speed"], CancellationToken.None);
+                if (drv.GetHealth().State == DriverState.Degraded) Interlocked.Increment(ref badHealth);
+                reads++;
+                started.TrySetResult();
+            }
+            return reads;
+        });
+        await started.Task.WaitAsync(TimeSpan.FromSeconds(5));
+
+        for (var i = 0; i < 200; i++)
+            await drv.InitializeAsync("{}", CancellationToken.None);
+        stop.Cancel();
+
+        (await reader).ShouldBeGreaterThan(0); // positive control: the reader actually ran
+        Volatile.Read(ref badHealth).ShouldBe(0, "a re-init is not a comms fault");
+        await drv.DisposeAsync();
+        created.Where(t => !t.Disposed).ShouldBeEmpty("every runtime a reader created must be disposed by some teardown");
+    }
+
+    /// <summary>F10: DisposeAsync is the complete, idempotent teardown (the runtime prefers it).</summary>
+    [Fact]
+    public async Task DisposeAsync_is_a_complete_idempotent_teardown()
+    {
+        var (drv, factory) = Build(probeReadDelay: TimeSpan.Zero);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        await WaitUntilAsync(() => factory.Tags.ContainsKey("ProbeTag"), "probe never started");
+        var loop = drv.GetDeviceState(Host)!.ProbeTask!;
+
+        await drv.DisposeAsync();
+        await drv.DisposeAsync(); // idempotent
+
+        loop.IsCompleted.ShouldBeTrue();
+        factory.Tags["ProbeTag"].Disposed.ShouldBeTrue();
+    }
+
+    /// <summary>
+    ///     F10: the sync <c>Dispose()</c> no longer blocks the caller's thread on the async teardown (sync-over-async —
+    ///     a dispatcher thread for up to the probe join). It starts the same teardown and returns; the teardown still
+    ///     completes.
+    /// </summary>
+    [Fact]
+    public async Task Sync_Dispose_does_not_block_on_the_teardown_but_still_completes_it()
+    {
+        var (drv, factory) = Build(probeReadDelay: TimeSpan.FromSeconds(4));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        await WaitUntilAsync(() => factory.Tags.ContainsKey("ProbeTag"), "probe never started");
+        var loop = drv.GetDeviceState(Host)!.ProbeTask!;
+        await Task.Delay(50); // the probe is inside its slow, cancellation-deaf read
+
+        var sw = System.Diagnostics.Stopwatch.StartNew();
+        ((IDisposable)drv).Dispose();
+        sw.Stop();
+
+        sw.Elapsed.ShouldBeLessThan(TimeSpan.FromSeconds(2)); // did not wait out the 4 s probe read (CI-tolerant margin)
+        await WaitUntilAsync(() => loop.IsCompleted && factory.Tags["ProbeTag"].Disposed, "teardown never completed", seconds: 15);
+        drv.GetDeviceState(Host).ShouldBeNull();
+    }
+
+    /// <summary>
+    ///     M4: a ShutdownAsync overtaken by a NEWER InitializeAsync (its probe join is slow) must not stamp Unknown over
+    ///     the newer init's Healthy when it finally finishes.
+    /// </summary>
+    [Fact]
+    public async Task A_slow_shutdown_does_not_overwrite_a_newer_inits_health()
+    {
+        var (drv, factory) = Build(probeReadDelay: TimeSpan.FromMilliseconds(400));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        await WaitUntilAsync(() => factory.Tags.ContainsKey("ProbeTag"), "probe never started");
+        await Task.Delay(30); // inside the slow probe read
+
+        var shutdown = drv.ShutdownAsync(CancellationToken.None);
+        await Task.Delay(30);
+        await drv.InitializeAsync("{}", CancellationToken.None); // newer: publishes while the shutdown still joins
+        await shutdown;
+
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+        drv.GetDeviceState(Host).ShouldNotBeNull();
+        await drv.DisposeAsync();
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipLocalRefusalTests.cs

@@ -7,7 +7,8 @@ namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
 
 /// <summary>
 ///     Task 22 review r1: a runtime that refuses an operation with <see cref="AbStatus.Unsupported"/> (the native
-///     transport does this before the wire, e.g. a write it cannot type) reaches the client as BadNotSupported —
+///     transport does this before the wire, e.g. a write it cannot type — an <see cref="AbCipLocalRefusalException"/>
+///     since #737 I2) reaches the client as BadNotSupported —
 ///     NOT as a comms fault. The handle is kept (one create, never disposed) and driver health is untouched.
 /// </summary>
 [Trait("Category", "Unit")]
@@ -19,12 +20,12 @@ public sealed class AbCipLocalRefusalTests
     {
         public override Task ReadAsync(CancellationToken cancellationToken) =>
             refuseRead
-                ? Task.FromException(new AbTransportException(AbStatus.Unsupported, "refused read"))
+                ? Task.FromException(new AbCipLocalRefusalException("refused read"))
                 : base.ReadAsync(cancellationToken);
 
         public override Task WriteAsync(CancellationToken cancellationToken) =>
             refuseWrite
-                ? Task.FromException(new AbTransportException(AbStatus.Unsupported, "refused write"))
+                ? Task.FromException(new AbCipLocalRefusalException("refused write"))
                 : base.WriteAsync(cancellationToken);
     }
 

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipProbeRefusalTests.cs

@@ -0,0 +1,86 @@
+using Microsoft.Extensions.Logging;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+using ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport;
+using CapturingLogger = ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests.AbCipLoggingTests.CapturingLogger<ZB.MOM.WW.OtOpcUa.Driver.AbCip.AbCipDriver>;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Review 07 H10: a typo'd (or download-deleted) <c>Probe.ProbeTagPath</c> used to record a Forward-Open
+///     FAILURE every probe tick, so the device's backoff window never closed and every cold / evicted data tag read
+///     <c>BadCommunicationError</c> forever — a permanent device blackout that a restart did not cure. The controller
+///     ANSWERED, so the probe now counts it as reachable: backoff success, host Running, one config Warning.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipProbeRefusalTests
+{
+    private const string Device = "ab://10.0.0.5/1,0";
+
+    private static (AbCipDriver Driver, CapturingLogger Logger) Build(Exception probeFault)
+    {
+        var logger = new CapturingLogger();
+        var factory = new FakeAbCipTagFactory
+        {
+            Customise = p => p.TagName == "ProbeTagTypo"
+                ? new FakeAbCipTag(p) { ThrowOnInitialize = true, Exception = probeFault }
+                : new FakeAbCipTag(p) { Value = 7 },
+        };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(new AbCipTagDefinition("Speed", Device, "Motor1.Speed", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions
+            {
+                Enabled = true,
+                ProbeTagPath = "ProbeTagTypo",
+                Interval = TimeSpan.FromMilliseconds(10),
+                Timeout = TimeSpan.FromMilliseconds(100),
+            },
+        }, "abcip-probe-refusal", factory, logger: logger);
+        return (drv, logger);
+    }
+
+    private static async Task WaitUntilAsync(Func<bool> condition)
+    {
+        var deadline = DateTime.UtcNow.AddSeconds(5);
+        while (!condition())
+        {
+            if (DateTime.UtcNow > deadline) throw new TimeoutException("condition never met");
+            await Task.Delay(5);
+        }
+    }
+
+    /// <summary>A NotFound probe tag leaves the device reachable: data tags still connect, host is Running, warned once.</summary>
+    [Fact]
+    public async Task A_missing_probe_tag_does_not_black_out_the_device()
+    {
+        var (drv, logger) = Build(new AbTransportException(AbStatus.NotFound, "NotFound"));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        await WaitUntilAsync(() => drv.GetHostStatuses().Single().State == HostState.Running);
+        await Task.Delay(100); // many probe ticks
+
+        var r = await drv.ReadAsync(["Speed"], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good); // a cold create, NOT throttled
+        r[0].Value.ShouldBe(7);
+        drv.GetHostStatuses().Single().State.ShouldBe(HostState.Running);
+        logger.Entries.Count(e => e.Level == LogLevel.Warning && e.Message.Contains("ProbeTagTypo")).ShouldBe(1);
+        await drv.DisposeAsync();
+    }
+
+    /// <summary>Positive control: a probe that genuinely cannot connect still stops the host and throttles cold creates.</summary>
+    [Fact]
+    public async Task An_unreachable_device_still_stops_the_host_and_throttles()
+    {
+        var (drv, _) = Build(new AbTransportException(AbStatus.Timeout, "Timeout"));
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        await WaitUntilAsync(() => drv.GetHostStatuses().Single().State == HostState.Stopped);
+
+        var r = await drv.ReadAsync(["Speed"], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadCommunicationError); // throttled by the open window
+        await drv.DisposeAsync();
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipReadWindowTests.cs

@@ -0,0 +1,96 @@
+using Shouldly;
+using Xunit;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Review 07 P8 / P7: the bounded-parallel read window. On the native transport every concurrent tag read of a
+///     device rides ONE shared request scheduler that packs whatever is queued into a Multiple Service Packet, so a
+///     window of 8 capped MSP coalescing at 8 services per round trip; it is widened to at least 64 there (the
+///     scheduler's reply budget already prevents over-packing). And a cold create waiting on a device's connect gate
+///     must not occupy a window slot, or one dead device's backoff-expiry attempt stalls every other device's reads.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipReadWindowTests
+{
+    private static AbCipDriverOptions Options(int maxConcurrentReads, AbCipTransport? transport = null) => new()
+    {
+        Devices = [new AbCipDeviceOptions("ab://10.0.0.5/1,0")],
+        Probe = new AbCipProbeOptions { Enabled = false },
+        MaxConcurrentReads = maxConcurrentReads,
+        Transport = transport,
+    };
+
+    /// <summary>The window per transport: native ≥ 64; libplctag and injected factories honour the option exactly.</summary>
+    [Fact]
+    public void The_native_window_is_at_least_64_and_other_transports_keep_the_option()
+    {
+        new AbCipDriver(Options(8, AbCipTransport.Native), "a").ReadWindow.ShouldBe(AbCipDriver.NativeMinReadWindow);
+        AbCipDriver.NativeMinReadWindow.ShouldBeGreaterThanOrEqualTo(64);
+        new AbCipDriver(Options(1, AbCipTransport.Native), "a").ReadWindow.ShouldBe(AbCipDriver.NativeMinReadWindow);
+        new AbCipDriver(Options(200, AbCipTransport.Native), "a").ReadWindow.ShouldBe(200);
+        new AbCipDriver(Options(8), "a", defaultTransport: AbCipTransport.Native).ReadWindow.ShouldBe(AbCipDriver.NativeMinReadWindow);
+
+        // Positive controls: libplctag (per-handle concurrency) and an injected test factory keep the option.
+        new AbCipDriver(Options(8, AbCipTransport.Libplctag), "a").ReadWindow.ShouldBe(8);
+        new AbCipDriver(Options(0, AbCipTransport.Libplctag), "a").ReadWindow.ShouldBe(1);
+        new AbCipDriver(Options(3, AbCipTransport.Native), "a", new FakeAbCipTagFactory()).ReadWindow.ShouldBe(3);
+    }
+
+    /// <summary>A tag whose initialisation blocks until released (a cold connect to a slow / dead device).</summary>
+    private sealed class BlockingInitTag(AbCipTagCreateParams p, Task release) : FakeAbCipTag(p)
+    {
+        public override async Task InitializeAsync(CancellationToken cancellationToken)
+        {
+            await base.InitializeAsync(cancellationToken);
+            await release.WaitAsync(cancellationToken);
+        }
+    }
+
+    /// <summary>
+    ///     With a window of ONE, device B's blocked cold create (and its queued sibling) must not stop device A's tag
+    ///     from being read: cold waiters sit on B's connect gate, not in the window.
+    /// </summary>
+    [Fact]
+    public async Task A_blocked_cold_create_on_one_device_does_not_hold_the_window_against_another()
+    {
+        var release = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
+        var factory = new FakeAbCipTagFactory
+        {
+            Customise = p => p.Gateway == "10.0.0.6"
+                ? new BlockingInitTag(p, release.Task) { Value = 2 }
+                : new FakeAbCipTag(p) { Value = 1 },
+        };
+        const string a = "ab://10.0.0.5/1,0", b = "ab://10.0.0.6/1,0";
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(a), new AbCipDeviceOptions(b)],
+            RawTags = AbCipRawTags.From(
+                new AbCipTagDefinition("B1", b, "B1", AbCipDataType.DInt),
+                new AbCipTagDefinition("B2", b, "B2", AbCipDataType.DInt),
+                new AbCipTagDefinition("A1", a, "A1", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+            MaxConcurrentReads = 1,
+        }, "abcip-window", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+
+        var read = drv.ReadAsync(["B1", "B2", "A1"], CancellationToken.None);
+        var deadline = DateTime.UtcNow.AddSeconds(5);
+        while (!(factory.Tags.TryGetValue("A1", out var tag) && tag.ReadCount > 0))
+        {
+            if (DateTime.UtcNow > deadline)
+            {
+                release.TrySetResult();
+                await read;
+                throw new ShouldAssertException("device A's read waited behind device B's blocked cold create");
+            }
+            await Task.Delay(5);
+        }
+        read.IsCompleted.ShouldBeFalse(); // B is still blocked — A was read regardless
+
+        release.SetResult();
+        var r = await read;
+        r.Select(x => x.Value).ShouldBe([2, 2, 1]); // positive control: B's reads complete once it connects
+        await drv.DisposeAsync();
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipStatusMapperTests.cs

@@ -77,7 +77,7 @@ public sealed class AbCipStatusMapperTests
         Status.Ok => AbCipStatusMapper.Good,
         Status.Pending => AbCipStatusMapper.GoodMoreData,
         Status.ErrorTimeout => AbCipStatusMapper.BadTimeout,
-        Status.ErrorNotFound or Status.ErrorNoMatch or Status.ErrorBadDevice => AbCipStatusMapper.BadNodeIdUnknown,
+        Status.ErrorNotFound => AbCipStatusMapper.BadNodeIdUnknown, // #737 I2: NoMatch/BadDevice are comms
         Status.ErrorNotAllowed => AbCipStatusMapper.BadNotWritable,
         Status.ErrorOutOfBounds or Status.ErrorTooLarge or Status.ErrorTooSmall => AbCipStatusMapper.BadOutOfRange,
         Status.ErrorUnsupported or Status.ErrorNotImplemented => AbCipStatusMapper.BadNotSupported,

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipTypeMismatchTests.cs

@@ -73,7 +73,7 @@ public sealed class AbCipTypeMismatchTests
         creates().ShouldBe(1);
         factory.Tags["T"].Disposed.ShouldBeFalse();
         factory.Tags["T"].ReadCount.ShouldBe(Polls); // positive control: every poll did reach the runtime
-        drv.GetHealth().ShouldBe(health);
+        AssertStateUnchanged(drv.GetHealth(), health);
         MismatchWarnings(log).ShouldBe(1);
     }
 
@@ -104,7 +104,7 @@ public sealed class AbCipTypeMismatchTests
         Volatile.Read(ref creates).ShouldBe(1);
         factory.Tags["T"].Disposed.ShouldBeFalse();
         factory.Tags["T"].ReadCount.ShouldBe(Polls); // positive control: every poll did reach the runtime
-        drv.GetHealth().ShouldBe(health);
+        AssertStateUnchanged(drv.GetHealth(), health);
         drv.GetHealth().State.ShouldNotBe(DriverState.Degraded);
         MismatchWarnings(log).ShouldBe(1);
     }
@@ -121,7 +121,7 @@ public sealed class AbCipTypeMismatchTests
 
         creates().ShouldBe(1);
         factory.Tags["T"].Disposed.ShouldBeFalse();
-        drv.GetHealth().ShouldBe(health);
+        AssertStateUnchanged(drv.GetHealth(), health);
         MismatchWarnings(log).ShouldBe(1);
     }
 
@@ -179,7 +179,7 @@ public sealed class AbCipTypeMismatchTests
             target.ForwardOpens.ShouldBe(1);
             target.ForwardCloses.ShouldBe(0);
             target.ConnectedRequests.ShouldBeGreaterThanOrEqualTo(Polls); // positive control: polls hit the wire
-            drv.GetHealth().ShouldBe(health);
+            AssertStateUnchanged(drv.GetHealth(), health);
             MismatchWarnings(log).ShouldBe(1);
         }
         finally
@@ -226,7 +226,7 @@ public sealed class AbCipTypeMismatchTests
             target.ForwardOpens.ShouldBe(1);
             target.ForwardCloses.ShouldBe(0);
             target.ConnectedRequests.ShouldBeGreaterThanOrEqualTo(Polls); // positive control: polls hit the wire
-            drv.GetHealth().ShouldBe(health);
+            AssertStateUnchanged(drv.GetHealth(), health);
         }
         finally
         {
@@ -267,12 +267,23 @@ public sealed class AbCipTypeMismatchTests
         Volatile.Read(ref creates).ShouldBe(1);
         factory.Tags["T"].Disposed.ShouldBeFalse();
         factory.Tags["T"].ReadCount.ShouldBe(Polls - 1); // positive control: later polls reached the kept runtime
-        drv.GetHealth().ShouldBe(health);
+        AssertStateUnchanged(drv.GetHealth(), health);
     }
 
     private sealed class CountingFactory(Func<AbCipTagCreateParams, IAbCipTagRuntime> create) : IAbCipTagFactory
     {
         public IAbCipTagRuntime Create(AbCipTagCreateParams createParams) => create(createParams);
     }
+    /// <summary>
+    ///     A refusal leaves the health STATE and error alone; since the controller answered, LastSuccessfulRead may
+    ///     only move forward (review 07 H1 — any data proving the device answered advances it).
+    /// </summary>
+    private static void AssertStateUnchanged(DriverHealth actual, DriverHealth before)
+    {
+        actual.State.ShouldBe(before.State);
+        actual.LastError.ShouldBe(before.LastError);
+        (actual.LastSuccessfulRead ?? DateTime.MinValue).ShouldBeGreaterThanOrEqualTo(before.LastSuccessfulRead ?? DateTime.MinValue);
+    }
+
 #pragma warning restore OTOPCUA0001
 }

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipUdtMemberLayoutTests.cs

@@ -99,4 +99,26 @@ public sealed class AbCipUdtMemberLayoutTests
             new AbCipStructureMember("Setpoints", AbCipDataType.Real, ElementCount: 4),
         }).ShouldBeNull();
     }
+
+    /// <summary>Review 07 P11: the layout is computed once per member list, not once per member per poll.</summary>
+    [Fact]
+    public void GetOrBuild_computes_a_layout_once_per_member_list()
+    {
+        IReadOnlyList<AbCipStructureMember> members =
+        [
+            new AbCipStructureMember("A", AbCipDataType.DInt),
+            new AbCipStructureMember("B", AbCipDataType.Real),
+        ];
+        var first = AbCipUdtMemberLayout.GetOrBuild(members);
+        first.ShouldNotBeNull();
+        first["B"].ShouldBe(4);
+        AbCipUdtMemberLayout.GetOrBuild(members).ShouldBeSameAs(first);
+
+        // Positive control: a different (equal-content) list is its own entry, and a not-groupable verdict caches too.
+        IReadOnlyList<AbCipStructureMember> copy = [.. members];
+        AbCipUdtMemberLayout.GetOrBuild(copy).ShouldNotBeSameAs(first);
+        IReadOnlyList<AbCipStructureMember> withBool = [new AbCipStructureMember("F", AbCipDataType.Bool)];
+        AbCipUdtMemberLayout.GetOrBuild(withBool).ShouldBeNull();
+        AbCipUdtMemberLayout.GetOrBuild(withBool).ShouldBeNull();
+    }
 }

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/AbCipWriteHealthTests.cs

@@ -0,0 +1,175 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Abstractions;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests;
+
+/// <summary>
+///     Review 07 H19: write-side health. A client value that cannot be converted never touched the wire, so it must
+///     not latch <c>Degraded</c>; and a good write must not overwrite a <c>Degraded</c> the read path just recorded
+///     with <c>Healthy</c> (the batch read path is worst-wins for exactly this reason). A good write proves the device
+///     answered, so it advances <c>LastSuccessfulRead</c> without touching the state.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class AbCipWriteHealthTests
+{
+    private const string Device = "ab://10.0.0.5/1,0";
+
+    private sealed class ThrowingWriteTag(AbCipTagCreateParams p) : FakeAbCipTag(p)
+    {
+        public override Task WriteAsync(CancellationToken cancellationToken) =>
+            throw new InvalidOperationException("socket reset");
+    }
+
+    private static async Task<(AbCipDriver Driver, FakeAbCipTagFactory Factory)> BuildAsync(
+        Func<AbCipTagCreateParams, FakeAbCipTag>? make = null)
+    {
+        var factory = new FakeAbCipTagFactory
+        {
+            Customise = make ?? (p => p.TagName == "Broken"
+                ? new FakeAbCipTag(p) { ThrowOnRead = true }
+                : new RealEncodingTag(p) { Value = 0 }),
+        };
+        var drv = new AbCipDriver(new AbCipDriverOptions
+        {
+            Devices = [new AbCipDeviceOptions(Device)],
+            RawTags = AbCipRawTags.From(
+                new AbCipTagDefinition("D", Device, "D", AbCipDataType.DInt),
+                new AbCipTagDefinition("I", Device, "I", AbCipDataType.Int),
+                new AbCipTagDefinition("Broken", Device, "Broken", AbCipDataType.DInt)),
+            Probe = new AbCipProbeOptions { Enabled = false },
+        }, "abcip-write-health", factory);
+        await drv.InitializeAsync("{}", CancellationToken.None);
+        return (drv, factory);
+    }
+
+    /// <summary>A fake whose encode converts like the real runtimes do (Convert.ToInt32 / ToInt16).</summary>
+    private sealed class RealEncodingTag(AbCipTagCreateParams p) : FakeAbCipTag(p)
+    {
+        public override void EncodeValue(AbCipDataType type, int? bitIndex, object? value) =>
+            Value = type switch
+            {
+                AbCipDataType.Int => Convert.ToInt16(value),
+                _ => Convert.ToInt32(value),
+            };
+    }
+
+    /// <summary>Unconvertible / out-of-range client values: refused per tag, health untouched, nothing sent.</summary>
+    [Fact]
+    public async Task A_client_conversion_error_writes_no_health()
+    {
+        var (drv, factory) = await BuildAsync();
+        var before = drv.GetHealth();
+
+        var r = await drv.WriteAsync(
+        [
+            new WriteRequest("D", "not a number"),   // FormatException
+            new WriteRequest("D", new object()),     // InvalidCastException
+            new WriteRequest("I", 1_000_000),        // OverflowException
+        ], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadTypeMismatch);
+        r[1].StatusCode.ShouldBe(AbCipStatusMapper.BadTypeMismatch);
+        r[2].StatusCode.ShouldBe(AbCipStatusMapper.BadOutOfRange);
+        drv.GetHealth().ShouldBe(before);
+        factory.Tags["D"].WriteCount.ShouldBe(0);
+        factory.Tags["I"].WriteCount.ShouldBe(0);
+    }
+
+    /// <summary>A good write keeps a Degraded the read path recorded, and advances LastSuccessfulRead.</summary>
+    [Fact]
+    public async Task A_good_write_does_not_overwrite_Degraded_with_Healthy()
+    {
+        var (drv, _) = await BuildAsync();
+        await drv.ReadAsync(["Broken"], CancellationToken.None);
+        var degraded = drv.GetHealth();
+        degraded.State.ShouldBe(DriverState.Degraded);
+        await Task.Delay(20);
+
+        var r = await drv.WriteAsync([new WriteRequest("D", 5)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.Good);
+        var after = drv.GetHealth();
+        after.State.ShouldBe(DriverState.Degraded);
+        after.LastError.ShouldBe(degraded.LastError);
+        after.LastSuccessfulRead!.Value.ShouldBeGreaterThan(degraded.LastSuccessfulRead!.Value); // the device answered
+    }
+
+    /// <summary>Positive control: a genuine write transport fault still degrades and preserves LastSuccessfulRead.</summary>
+    [Fact]
+    public async Task A_write_transport_fault_still_degrades()
+    {
+        var (drv, factory) = await BuildAsync(p => new ThrowingWriteTag(p));
+        var before = drv.GetHealth();
+
+        var r = await drv.WriteAsync([new WriteRequest("D", 5)], CancellationToken.None);
+
+        r[0].StatusCode.ShouldBe(AbCipStatusMapper.BadCommunicationError);
+        var after = drv.GetHealth();
+        after.State.ShouldBe(DriverState.Degraded);
+        after.LastSuccessfulRead.ShouldBe(before.LastSuccessfulRead);
+        factory.Tags["D"].Disposed.ShouldBeTrue();
+    }
+
+    /// <summary>M3: a good write clears a Degraded that a WRITE comms fault recorded.</summary>
+    [Fact]
+    public async Task A_good_write_clears_a_write_side_Degraded()
+    {
+        var fail = true;
+        var (drv, _) = await BuildAsync(p => fail ? new ThrowingWriteTag(p) : new RealEncodingTag(p));
+        (await drv.WriteAsync([new WriteRequest("D", 5)], CancellationToken.None))[0].StatusCode
+            .ShouldBe(AbCipStatusMapper.BadCommunicationError);
+        drv.GetHealth().State.ShouldBe(DriverState.Degraded);
+
+        fail = false; // the evicted runtime is re-created healthy
+        (await drv.WriteAsync([new WriteRequest("D", 6)], CancellationToken.None))[0].StatusCode
+            .ShouldBe(AbCipStatusMapper.Good);
+
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+        drv.GetHealth().LastError.ShouldBeNull();
+    }
+
+    /// <summary>
+    ///     m1: many concurrent failing writes leave the write-owned marker naming the installed Degraded, so one good
+    ///     write still clears it.
+    /// </summary>
+    [Fact]
+    public async Task Concurrent_failing_writes_then_a_good_write_clear_the_write_side_Degraded()
+    {
+        var fail = true;
+        var (drv, _) = await BuildAsync(p => fail ? new ThrowingWriteTag(p) : new RealEncodingTag(p));
+        for (var round = 0; round < 20; round++)
+        {
+            fail = true;
+            await Task.WhenAll(Enumerable.Range(0, 8).Select(_ => Task.Run(() =>
+                drv.WriteAsync([new WriteRequest("D", 1), new WriteRequest("I", 2)], CancellationToken.None))));
+            drv.GetHealth().State.ShouldBe(DriverState.Degraded);
+
+            fail = false;
+            (await drv.WriteAsync([new WriteRequest("D", 3)], CancellationToken.None))[0].StatusCode
+                .ShouldBe(AbCipStatusMapper.Good);
+            drv.GetHealth().State.ShouldBe(DriverState.Healthy, $"round {round}");
+        }
+    }
+
+    /// <summary>
+    ///     M3: a refusal-only read batch in between (no read comms fault happened) keeps the Degraded write-owned, so a
+    ///     good write still clears it.
+    /// </summary>
+    [Fact]
+    public async Task A_refusal_only_read_between_keeps_the_Degraded_write_owned()
+    {
+        var fail = true;
+        var (drv, _) = await BuildAsync(p => p.TagName == "Broken"
+            ? new FakeAbCipTag(p) { Status = (int)ZB.MOM.WW.OtOpcUa.Driver.Ab.Transport.AbStatus.NotFound }
+            : fail ? new ThrowingWriteTag(p) : new RealEncodingTag(p));
+        await drv.WriteAsync([new WriteRequest("D", 5)], CancellationToken.None);
+        drv.GetHealth().State.ShouldBe(DriverState.Degraded);
+        (await drv.ReadAsync(["Broken"], CancellationToken.None))[0].StatusCode.ShouldBe(AbCipStatusMapper.BadNodeIdUnknown);
+
+        fail = false;
+        await drv.WriteAsync([new WriteRequest("D", 6)], CancellationToken.None);
+
+        drv.GetHealth().State.ShouldBe(DriverState.Healthy);
+    }
+}

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/FakeAbCipTag.cs

@@ -115,6 +115,25 @@ internal class FakeAbCipTag : IAbCipTagRuntime
     /// <param name="value">The value to encode.</param>
     public virtual void EncodeValue(AbCipDataType type, int? bitIndex, object? value) => Value = value;
 
+    /// <summary>The element type of the last <see cref="EncodeArray"/>, if any.</summary>
+    public AbCipDataType? EncodedArrayType { get; private set; }
+
+    /// <summary>The bytes the last <see cref="EncodeArray"/> produced (the shared codec, so conversions throw as in production).</summary>
+    public byte[]? EncodedArrayBytes { get; private set; }
+
+    /// <summary>The element count of the last <see cref="EncodeArray"/>.</summary>
+    public int EncodedArrayCount { get; private set; }
+
+    /// <summary>Encodes an array write through the production codec and records it.</summary>
+    /// <param name="type">The element type.</param>
+    /// <param name="values">The element values.</param>
+    public virtual void EncodeArray(AbCipDataType type, IReadOnlyList<object?> values)
+    {
+        EncodedArrayBytes = AbCipArrayCodec.Encode(type, values);
+        EncodedArrayType = type;
+        EncodedArrayCount = values.Count;
+    }
+
     /// <summary>Marks the tag as disposed.</summary>
     public virtual void Dispose() => Disposed = true;
 }

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/LibplctagStatusTranslationTests.cs

@@ -16,8 +16,8 @@ public sealed class LibplctagStatusTranslationTests
     [InlineData(Status.Pending, AbStatus.Pending)]
     [InlineData(Status.ErrorTimeout, AbStatus.Timeout)]
     [InlineData(Status.ErrorNotFound, AbStatus.NotFound)]
-    [InlineData(Status.ErrorNoMatch, AbStatus.NotFound)]
-    [InlineData(Status.ErrorBadDevice, AbStatus.NotFound)]
+    [InlineData(Status.ErrorNoMatch, AbStatus.BadConnection)]   // #737 I2: not proof the controller answered
+    [InlineData(Status.ErrorBadDevice, AbStatus.BadConnection)]
     [InlineData(Status.ErrorNotAllowed, AbStatus.NotAllowed)]
     [InlineData(Status.ErrorOutOfBounds, AbStatus.OutOfBounds)]
     [InlineData(Status.ErrorTooLarge, AbStatus.TooLarge)]
@@ -148,4 +148,24 @@ public sealed class LibplctagStatusTranslationTests
             LibplctagStatusTranslation.Wrap(() => throw new LibPlcTagException(Status.ErrorBadParam)));
         thrown.Status.ShouldBe(AbStatus.BadConnection);
     }
+
+    /// <summary>
+    ///     #737 I2: libplctag's <c>ErrorNotImplemented</c> is the LIBRARY refusing (no I/O): a local refusal, not a device
+    ///     refusal — it must not count as "the controller answered".
+    /// </summary>
+    [Fact]
+    public async Task NotImplemented_is_a_local_refusal_not_a_device_refusal()
+    {
+        var ex = await Should.ThrowAsync<AbTransportException>(() =>
+            LibplctagStatusTranslation.Wrap(() => throw new LibPlcTagException("ErrorNotImplemented")));
+        ex.ShouldBeOfType<AbCipLocalRefusalException>();
+        ex.Status.ShouldBe(AbStatus.Unsupported);
+        AbCipRefusals.IsDeviceRefusal(ex).ShouldBeFalse();
+        AbCipDriverProbe.IsReachableException(ex).ShouldBeFalse();
+
+        // Positive control: the device's own Unsupported refusal still is one.
+        var device = await Should.ThrowAsync<AbTransportException>(() =>
+            LibplctagStatusTranslation.Wrap(() => throw new LibPlcTagException("ErrorUnsupported")));
+        AbCipRefusals.IsDeviceRefusal(device).ShouldBeTrue();
+    }
 }

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.AbCip.Tests/NativeAbCipTagRuntimeTests.cs

@@ -112,6 +112,163 @@ public sealed class NativeAbCipTagRuntimeTests
         rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(true);
     }
 
+    /// <summary>
+    ///     Review 07 U11: a scalar BOOL authored over a BOOL-array element <c>Bits[n]</c> reads DWORD <c>n / 32</c>
+    ///     (a Logix BOOL array is stored as DWORDs and its CIP element index addresses them) and decodes bit
+    ///     <c>n % 32</c> of it — it used to request element n and report "any of bits 0-7 of the reply set".
+    /// </summary>
+    /// <param name="n">The authored BOOL-array element.</param>
+    /// <param name="dword">The DWORD the request must address.</param>
+    /// <param name="reply">The DWORD the controller returns (little-endian hex).</param>
+    /// <param name="expected">The element's value.</param>
+    [Theory]
+    [InlineData(37, 1, "20000000", true)]   // bit 5 of DWORD 1
+    [InlineData(36, 1, "20000000", false)]  // bit 4 of the same DWORD — the old "low byte != 0" read true
+    [InlineData(5, 0, "20000000", true)]
+    [InlineData(1, 0, "01000000", false)]   // bit 1, not bit 0
+    [InlineData(31, 0, "00000080", true)]   // top bit of DWORD 0 — outside the low byte entirely
+    public async Task A_BOOL_array_element_decodes_its_bit_of_its_DWORD(int n, int dword, string reply, bool expected)
+    {
+        var ch = Replying("D300", reply);
+        using var rt = await ReadAsync(ch, Params($"Bits[{n}]", AbCipDataType.Bool));
+        rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(expected);
+        ch.Sent[0].ShouldBe(Convert.FromHexString("5204" + "910442697473" + "28" + dword.ToString("X2") + "0100" + "00000000"));
+    }
+
+    /// <summary>
+    ///     #737 I1: a family whose BOOL array is element-addressed (the controller answers the DWORD-index request with a
+    ///     plain BOOL, 0xC1) falls back to the ORIGINAL element path <c>Bits[n]</c> — it must neither decode nor write
+    ///     element <c>n / 32</c>. Positive control: the 0xD3 theory above still decodes bit <c>n % 32</c>.
+    /// </summary>
+    [Fact]
+    public async Task A_per_element_BOOL_array_falls_back_to_the_element_path_for_read_and_write()
+    {
+        var ch = new FakeCipChannel
+        {
+            Respond = req => Task.FromResult(Convert.FromHexString(req[0] == 0x4D ? "CD000000" : "D2000000C10001")),
+        };
+        using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Bits[37]", AbCipDataType.Bool));
+        await rt.InitializeAsync(CancellationToken.None);
+        await rt.ReadAsync(CancellationToken.None);
+
+        rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(true);
+        ch.Sent[^1].ShouldBe(Convert.FromHexString("5204" + "910442697473" + "2825" + "0100" + "00000000")); // Bits[37]
+
+        rt.EncodeValue(AbCipDataType.Bool, null, false);
+        await rt.WriteAsync(CancellationToken.None);
+        ch.Sent[^1].ShouldBe(Convert.FromHexString("4D04" + "910442697473" + "2825" + "C100" + "0100" + "00")); // Bits[37]
+    }
+
+    /// <summary>A channel whose reply can be switched mid-test (read replies; writes always succeed).</summary>
+    private static FakeCipChannel Switchable(Func<string> readReply) => new()
+    {
+        Respond = req => Task.FromResult(Convert.FromHexString(req[0] == 0x4D ? "CD000000" : readReply())),
+    };
+
+    private const string Dword1 = "5204" + "910442697473" + "2801" + "0100" + "00000000";   // Bits[1]  (DWORD 37/32)
+    private const string Element37 = "5204" + "910442697473" + "2825" + "0100" + "00000000"; // Bits[37] (element)
+
+    /// <summary>
+    ///     #737 fix round 2 (N1): a REFUSAL of the DWORD request says nothing about the layout — it must not settle it.
+    ///     Tag deployed before the online edit that creates <c>Bits : BOOL[2048]</c>: the init read is refused NotFound;
+    ///     once the array exists the controller answers 0xD3, and the read must address DWORD n/32 and decode bit n%32
+    ///     (the round-1 code had permanently fallen back to <c>Bits[37]</c> = DWORD 37 and decoded byte 0 as Good).
+    /// </summary>
+    /// <param name="refusal">The refusal reply the DWORD request first gets.</param>
+    [Theory]
+    [InlineData("D2000400")]   // NotFound
+    [InlineData("D2000F00")]   // NotAllowed
+    [InlineData("D2000A00")]   // OutOfBounds
+    public async Task A_refusal_leaves_the_BOOL_layout_unsettled_and_a_later_DWORD_reply_decodes_the_right_bit(string refusal)
+    {
+        var reply = refusal;
+        var ch = Switchable(() => reply);
+        using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Bits[37]", AbCipDataType.Bool));
+        await rt.InitializeAsync(CancellationToken.None); // refused: completes with the refusal status
+        rt.GetStatus().ShouldNotBe((int)AbStatus.Ok);
+        ch.Sent[^1].ShouldBe(Convert.FromHexString(Dword1)); // never fell back to the element path
+
+        reply = "D2000000D300" + "20000000"; // bit 5 of DWORD 1 = element 37
+        await rt.ReadAsync(CancellationToken.None);
+
+        ch.Sent[^1].ShouldBe(Convert.FromHexString(Dword1));
+        rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(true);
+    }
+
+    /// <summary>
+    ///     N1 positive control: a refusal, then a 0xC1 reply, settles PER-ELEMENT — the n/32 element that came back is
+    ///     discarded and <c>Bits[n]</c> is re-read and decoded.
+    /// </summary>
+    [Fact]
+    public async Task A_refusal_then_a_per_element_reply_settles_the_element_path()
+    {
+        var reply = "D2000A00";
+        var ch = Switchable(() => reply);
+        using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Bits[37]", AbCipDataType.Bool));
+        await rt.InitializeAsync(CancellationToken.None);
+
+        reply = "D2000000C10001";
+        await rt.ReadAsync(CancellationToken.None);
+
+        ch.Sent[^2].ShouldBe(Convert.FromHexString(Dword1));     // asked DWORD-addressed, got a plain BOOL…
+        ch.Sent[^1].ShouldBe(Convert.FromHexString(Element37));  // …so element 37 is re-read and decoded
+        rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(true);
+    }
+
+    /// <summary>
+    ///     N1: a layout that changes after it was settled (an online edit re-types the array) re-settles from the reply
+    ///     type — it never decodes a DWORD reply to the ELEMENT request (DWORD n, byte 0) as the element's value.
+    /// </summary>
+    [Fact]
+    public async Task A_settled_per_element_layout_resettles_when_the_reply_becomes_a_DWORD()
+    {
+        var reply = "D2000000C10000";
+        var ch = Switchable(() => reply);
+        using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Bits[36]", AbCipDataType.Bool));
+        await rt.InitializeAsync(CancellationToken.None); // settles per-element
+
+        reply = "D2000000D300" + "01000000"; // byte 0 non-zero, bit 4 (element 36) clear
+        await rt.ReadAsync(CancellationToken.None);
+
+        ch.Sent[^1].ShouldBe(Convert.FromHexString("5204" + "910442697473" + "2801" + "0100" + "00000000"));
+        rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(false); // bit 4 of DWORD 1, not "byte 0 != 0"
+    }
+
+    /// <summary>N1 minimum guard: a scalar BOOL with no element bit answered as a packed DWORD is refused, not byte 0.</summary>
+    [Fact]
+    public async Task A_scalar_BOOL_answered_as_a_packed_DWORD_without_an_element_bit_is_refused()
+    {
+        using var rt = await ReadAsync(Replying("D300", "01000000"), Params("Flag", AbCipDataType.Bool));
+        (Should.Throw<AbTransportException>(() => rt.DecodeValue(AbCipDataType.Bool, null))).Status.ShouldBe(AbStatus.TypeMismatch);
+    }
+
+    /// <summary>
+    ///     m4: a runtime whose init read failed on the wire (layout still unsettled) settles the layout by a read BEFORE
+    ///     it writes, so the write goes to the right element of a per-element array.
+    /// </summary>
+    [Fact]
+    public async Task A_write_on_an_unsettled_layout_reads_first_and_writes_the_element()
+    {
+        var reply = "D2000100"; // 0x01: not a refusal — init throws, layout unsettled
+        var ch = Switchable(() => reply);
+        using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Bits[37]", AbCipDataType.Bool));
+        await Should.ThrowAsync<AbTransportException>(() => rt.InitializeAsync(CancellationToken.None));
+
+        reply = "D2000000C10000";
+        rt.EncodeValue(AbCipDataType.Bool, null, true);
+        await rt.WriteAsync(CancellationToken.None);
+
+        ch.Sent[^1].ShouldBe(Convert.FromHexString("4D04" + "910442697473" + "2825" + "C100" + "0100" + "01")); // Bits[37]
+    }
+
+    /// <summary>Positive control: a BOOL answered as a plain BOOL (0xC1) still decodes its byte.</summary>
+    [Fact]
+    public async Task A_BOOL_element_answered_as_BOOL_decodes_the_byte()
+    {
+        using var rt = await ReadAsync(Replying("C100", "01"), Params("Flag", AbCipDataType.Bool));
+        rt.DecodeValue(AbCipDataType.Bool, null).ShouldBe(true);
+    }
+
     [Fact]
     public async Task Real_returned_for_a_DInt_authored_tag_is_TypeMismatch()
     {
@@ -129,14 +286,35 @@ public sealed class NativeAbCipTagRuntimeTests
         rt.GetStatus().ShouldBe((int)AbStatus.Ok);
     }
 
-    [Fact]
-    public async Task Unknown_tag_fails_initialize_with_NotFound()
+    /// <summary>
+    ///     Review 07 L4: a controller refusal on the existence read completes the initialization (the session works —
+    ///     throwing made the driver dispose + re-create the runtime, re-leasing the connection, every poll); the
+    ///     refusal stays readable and every later read reports it again.
+    /// </summary>
+    [Theory]
+    [InlineData("D2000400", AbStatus.NotFound)]
+    [InlineData("D2000500", AbStatus.NotFound)]
+    [InlineData("D2000A00", AbStatus.OutOfBounds)]
+    [InlineData("D2000F00", AbStatus.NotAllowed)]
+    [InlineData("D200FF010521", AbStatus.OutOfBounds)]
+    public async Task A_refused_tag_completes_initialize_and_reports_the_refusal_on_read(string reply, AbStatus expected)
     {
-        var ch = new FakeCipChannel { Respond = _ => Task.FromResult(Convert.FromHexString("D2000400")) };
+        var ch = new FakeCipChannel { Respond = _ => Task.FromResult(Convert.FromHexString(reply)) };
+        using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Nope", AbCipDataType.DInt));
+        await rt.InitializeAsync(CancellationToken.None);
+        rt.GetStatus().ShouldBe((int)expected);
+        (await Should.ThrowAsync<AbTransportException>(() => rt.ReadAsync(CancellationToken.None)))
+            .Status.ShouldBe(expected);
+    }
+
+    /// <summary>Positive control: a reply the transport cannot classify as a refusal still fails initialize.</summary>
+    [Fact]
+    public async Task An_unclassified_reply_still_fails_initialize()
+    {
+        var ch = new FakeCipChannel { Respond = _ => Task.FromResult(Convert.FromHexString("D2000100")) };
         using var rt = NativeAbCipTagRuntime.ForTest(ch, Params("Nope", AbCipDataType.DInt));
         (await Should.ThrowAsync<AbTransportException>(() => rt.InitializeAsync(CancellationToken.None)))
-            .Status.ShouldBe(AbStatus.NotFound);
-        rt.GetStatus().ShouldBe((int)AbStatus.NotFound);
+            .Status.ShouldBe(AbStatus.BadReply);
     }
 
     [Fact]

tests/Drivers/ZB.MOM.WW.OtOpcUa.Driver.Calculation.Tests/CalculationScriptingOptionsResolutionTests.cs

@@ -0,0 +1,69 @@
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Scripting;
+
+namespace ZB.MOM.WW.OtOpcUa.Driver.Calculation.Tests;
+
+/// <summary>
+///     The Calculation driver's run-guard options: the node-wide <c>Scripting</c> section is the base, and a key
+///     authored in the driver's own config overrides only that knob. Before the section was bound, a Calculation
+///     driver always ran on the compiled defaults whatever the node configured.
+/// </summary>
+[Trait("Category", "Unit")]
+public sealed class CalculationScriptingOptionsResolutionTests
+{
+    private static readonly ScriptingOptions Node = new()
+    {
+        MaxAllocBytesPerRun = 8L * 1024 * 1024,
+        QuarantineAfterConsecutiveTimeouts = 7,
+        StuckOrphanQuarantineFactor = 4,
+    };
+
+    [Fact]
+    public void No_driver_keys_and_no_node_options_resolves_null_so_the_guard_uses_its_defaults() =>
+        CalculationDriverFactoryExtensions.ResolveScripting(null, null, null, out _).ShouldBeNull();
+
+    [Fact]
+    public void No_driver_keys_inherits_every_node_knob()
+    {
+        var resolved = CalculationDriverFactoryExtensions.ResolveScripting(null, null, Node, out _).ShouldNotBeNull();
+
+        resolved.MaxAllocBytesPerRun.ShouldBe(Node.MaxAllocBytesPerRun);
+        resolved.QuarantineAfterConsecutiveTimeouts.ShouldBe(7);
+        resolved.StuckOrphanQuarantineFactor.ShouldBe(4);
+    }
+
+    [Fact]
+    public void A_driver_key_overrides_only_its_own_knob_and_the_node_object_is_not_mutated()
+    {
+        var resolved = CalculationDriverFactoryExtensions.ResolveScripting(null, 2, Node, out var adjustments).ShouldNotBeNull();
+
+        resolved.QuarantineAfterConsecutiveTimeouts.ShouldBe(2);
+        resolved.MaxAllocBytesPerRun.ShouldBe(Node.MaxAllocBytesPerRun, "the unauthored key keeps the node value");
+        resolved.StuckOrphanQuarantineFactor.ShouldBe(4, "the stuck factor has no driver key; it is node-wide");
+        resolved.ShouldNotBeSameAs(Node);
+        Node.QuarantineAfterConsecutiveTimeouts.ShouldBe(7);
+        adjustments.ShouldBeEmpty();
+    }
+
+    /// <summary>Review m-2: an out-of-range value authored in the driver's own config is clamped AND reported, so
+    /// the factory can log it rather than the guard changing it silently.</summary>
+    [Fact]
+    public void An_out_of_range_driver_key_is_clamped_and_reported()
+    {
+        var resolved = CalculationDriverFactoryExtensions.ResolveScripting(1024L, 0, Node, out var adjustments).ShouldNotBeNull();
+
+        resolved.MaxAllocBytesPerRun.ShouldBe(ScriptingOptions.MinMaxAllocBytesPerRun);
+        resolved.QuarantineAfterConsecutiveTimeouts.ShouldBe(1);
+        adjustments.Count.ShouldBe(2);
+    }
+
+    [Fact]
+    public void A_driver_key_with_no_node_options_layers_on_the_compiled_defaults()
+    {
+        var resolved = CalculationDriverFactoryExtensions.ResolveScripting(0L, null, null, out _).ShouldNotBeNull();
+
+        resolved.MaxAllocBytesPerRun.ShouldBe(0);
+        resolved.QuarantineAfterConsecutiveTimeouts.ShouldBe(ScriptingOptions.DefaultQuarantineAfterConsecutiveTimeouts);
+    }
+}

tests/Server/ZB.MOM.WW.OtOpcUa.AdminUI.Tests/Authorization/PageAuthorizationGuardTests.cs

@@ -57,10 +57,11 @@ public class PageAuthorizationGuardTests
         //    AddSecretsAuthorization(). ──
         [typeof(SecretsAdmin)] = global::ZB.MOM.WW.Secrets.Ui.SecretsAuthorization.ManagePolicy,
 
-        // ── AuthenticatedRead (16): dashboards, lists, live tails (read-only) + the dev ContextMenu demo ──
+        // ── AuthenticatedRead (17): dashboards, lists, live tails (read-only) + the dev ContextMenu demo ──
         [typeof(Home)] = AdminUiPolicies.AuthenticatedRead,
         [typeof(global::ZB.MOM.WW.OtOpcUa.AdminUI.Components.Pages.Dev.ContextMenuDemo)] = AdminUiPolicies.AuthenticatedRead,
         [typeof(Fleet)] = AdminUiPolicies.AuthenticatedRead,
+        [typeof(global::ZB.MOM.WW.OtOpcUa.AdminUI.Components.Pages.FleetDiagnostics)] = AdminUiPolicies.AuthenticatedRead,
         [typeof(global::ZB.MOM.WW.OtOpcUa.AdminUI.Components.Pages.Hosts)] = AdminUiPolicies.AuthenticatedRead,
         [typeof(Alerts)] = AdminUiPolicies.AuthenticatedRead,
         [typeof(ScriptLog)] = AdminUiPolicies.AuthenticatedRead,

tests/Server/ZB.MOM.WW.OtOpcUa.AdminUI.Tests/ScriptAnalysis/CompletionTests.cs

@@ -18,6 +18,7 @@ public sealed class CompletionTests
         labels.ShouldContain("SetVirtualTag");
         labels.ShouldContain("Now");
         labels.ShouldContain("Logger");
+        labels.ShouldContain("CancellationToken");
     }
 
     [Fact] public async Task Scope_completion_includes_the_ctx_local()

tests/Server/ZB.MOM.WW.OtOpcUa.AdminUI.Tests/ScriptAnalysis/DiagnoseTests.cs

@@ -48,6 +48,7 @@ public sealed class DiagnoseTests
     [InlineData("""var x = 41; System.Threading.Interlocked.Increment(ref x); return x;""")]
     [InlineData("""var x = 42; return System.Threading.Volatile.Read(ref x);""")]
     [InlineData("""object o = 42; return o.GetType().IsValueType ? 1 : 0;""")]
+    [InlineData("""ctx.CancellationToken.ThrowIfCancellationRequested(); return ctx.CancellationToken.IsCancellationRequested ? 1 : 0;""")]
     public void Allow_listed_forms_are_not_flagged_in_the_editor(string code)
         => Diag(code).ShouldBeEmpty();
 

tests/Server/ZB.MOM.WW.OtOpcUa.Host.Tests/Configuration/ScriptingOptionsBindingTests.cs

@@ -0,0 +1,64 @@
+using Microsoft.Extensions.Configuration;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Core.Scripting;
+using ZB.MOM.WW.OtOpcUa.Host.Configuration;
+
+namespace ZB.MOM.WW.OtOpcUa.Host.Tests.Configuration;
+
+/// <summary>
+///     Review m-1: the <c>Scripting</c> section binds by the documented key names and is clamped at bind. Each
+///     configured value differs from the compiled default, so a misspelled key (which the binder would silently
+///     ignore, falling back to the default) goes red.
+/// </summary>
+public sealed class ScriptingOptionsBindingTests
+{
+    private static IConfiguration Config(Dictionary<string, string?> values) =>
+        new ConfigurationBuilder().AddInMemoryCollection(values).Build();
+
+    [Fact]
+    public void Every_documented_key_binds_to_its_property()
+    {
+        var bound = ScriptingOptionsBinding.Bind(Config(new()
+        {
+            ["Scripting:MaxAllocBytesPerRun"] = "8388608",
+            ["Scripting:QuarantineAfterConsecutiveTimeouts"] = "7",
+            ["Scripting:StuckOrphanQuarantineFactor"] = "4.5",
+            // Sibling keys owned by the script-log pipeline must not disturb the bind.
+            ["Scripting:LogFilePath"] = "logs/x-.log",
+            ["Scripting:LogTopicMinLevel"] = "Warning",
+        }), out var adjustments);
+
+        adjustments.ShouldBeEmpty();
+        bound.MaxAllocBytesPerRun.ShouldBe(8L * 1024 * 1024);
+        bound.QuarantineAfterConsecutiveTimeouts.ShouldBe(7);
+        bound.StuckOrphanQuarantineFactor.ShouldBe(4.5);
+    }
+
+    [Fact]
+    public void Out_of_range_values_are_clamped_at_bind_and_reported()
+    {
+        var bound = ScriptingOptionsBinding.Bind(Config(new()
+        {
+            ["Scripting:StuckOrphanQuarantineFactor"] = "1e300",
+            ["Scripting:MaxAllocBytesPerRun"] = "1024",
+            ["Scripting:QuarantineAfterConsecutiveTimeouts"] = "0",
+        }), out var adjustments);
+
+        bound.StuckOrphanQuarantineFactor.ShouldBe(ScriptingOptions.MaxStuckOrphanQuarantineFactor);
+        bound.MaxAllocBytesPerRun.ShouldBe(ScriptingOptions.MinMaxAllocBytesPerRun);
+        bound.QuarantineAfterConsecutiveTimeouts.ShouldBe(1);
+        adjustments.Count.ShouldBe(3);
+    }
+
+    [Fact]
+    public void An_absent_section_yields_the_compiled_defaults()
+    {
+        var bound = ScriptingOptionsBinding.Bind(Config(new()), out var adjustments);
+
+        adjustments.ShouldBeEmpty();
+        bound.MaxAllocBytesPerRun.ShouldBe(ScriptingOptions.DefaultMaxAllocBytesPerRun);
+        bound.QuarantineAfterConsecutiveTimeouts.ShouldBe(ScriptingOptions.DefaultQuarantineAfterConsecutiveTimeouts);
+        bound.StuckOrphanQuarantineFactor.ShouldBe(ScriptingOptions.DefaultStuckOrphanQuarantineFactor);
+    }
+}

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/Drivers/DriverHostActorScriptedAlarmRestartTests.cs

@@ -0,0 +1,103 @@
+using System.Text.Json;
+using Akka.Actor;
+using Serilog;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Commons.Engines;
+using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;
+using ZB.MOM.WW.OtOpcUa.Commons.Types;
+using ZB.MOM.WW.OtOpcUa.Configuration.Entities;
+using ZB.MOM.WW.OtOpcUa.Configuration.Enums;
+using ZB.MOM.WW.OtOpcUa.Core.ScriptedAlarms;
+using ZB.MOM.WW.OtOpcUa.Core.Scripting;
+using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
+using ZB.MOM.WW.OtOpcUa.Runtime.OpcUa;
+using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
+using ZB.MOM.WW.OtOpcUa.Runtime.VirtualTags;
+
+namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
+
+/// <summary>
+///     Review I-2: under the factory Props a supervision restart of the scripted-alarm host builds a fresh, EMPTY
+///     engine. Nothing re-sent <c>ApplyScriptedAlarms</c>, so until the next deploy the host had no mux interest
+///     (inputs stopped) and its ownership filter dropped every operator ack/shelve as "not owned". The host now
+///     tells its parent <see cref="ScriptedAlarms.ScriptedAlarmHostActor.ScriptedAlarmHostRestarted"/> from <c>PostRestart</c>,
+///     and <see cref="DriverHostActor"/> re-sends the last plans it applied. This test never re-applies by hand.
+/// </summary>
+public sealed class DriverHostActorScriptedAlarmRestartTests : RuntimeActorTestBase
+{
+    private static readonly NodeId TestNode = NodeId.Parse("driver-alarmhost-restart-test");
+    private static readonly RevisionHash RevA = RevisionHash.Parse(new string('d', 64));
+    private static readonly TimeSpan Timeout = TimeSpan.FromSeconds(30);
+
+    private static DeploymentId SeedAlarmDeployment(Microsoft.EntityFrameworkCore.IDbContextFactory<ZB.MOM.WW.OtOpcUa.Configuration.OtOpcUaConfigDbContext> db)
+    {
+        var artifact = JsonSerializer.SerializeToUtf8Bytes(new
+        {
+            ScriptedAlarms = new[]
+            {
+                new
+                {
+                    ScriptedAlarmId = "alm-1", EquipmentId = "eq-1", Name = "HighTemp", AlarmType = "AlarmCondition",
+                    Severity = 800, MessageTemplate = "hot", PredicateScriptId = "s-1",
+                    HistorizeToAveva = false, Retain = true, Enabled = true,
+                },
+            },
+            Scripts = new[] { new { ScriptId = "s-1", SourceCode = "return (int)ctx.GetTag(\"M.T\").Value > 90;" } },
+        });
+        var id = DeploymentId.NewId();
+        using var ctx = db.CreateDbContext();
+        ctx.Deployments.Add(new Deployment
+        {
+            DeploymentId = id.Value, RevisionHash = RevA.Value, Status = DeploymentStatus.Sealed,
+            CreatedBy = "test", SealedAtUtc = DateTime.UtcNow, ArtifactBlob = artifact,
+        });
+        ctx.SaveChanges();
+        return id;
+    }
+
+    [Fact]
+    public void After_a_supervision_restart_the_alarm_host_is_re_applied_and_inputs_and_acks_work_again()
+    {
+        var db = NewInMemoryDbFactory();
+        var deploymentId = SeedAlarmDeployment(db);
+        var publish = this.CreateCleanOutcomePublishProbe();
+        var mux = CreateTestProbe();
+        var coordinator = CreateTestProbe();
+        var host = Sys.ActorOf(DriverHostActor.Props(
+            db, TestNode, coordinator.Ref,
+            localRoles: new HashSet<string> { "driver" },
+            dependencyMux: mux.Ref,
+            opcUaPublishActor: publish.Ref,
+            virtualTagEvaluator: NullVirtualTagEvaluator.Instance,
+            virtualTagHostOverride: CreateTestProbe().Ref,
+            scriptRootLogger: new ScriptRootLogger(new LoggerConfiguration().CreateLogger()),
+            alarmStateStore: new InMemoryAlarmStateStore()));
+        V3ArtifactSeeder.Dispatch(host, coordinator, deploymentId, RevA);
+
+        var first = mux.FishForMessage<DependencyMuxActor.RegisterInterest>(r => r.TagRefs.Contains("M.T"), Timeout);
+        var alarmHost = first.Subscriber;
+
+        // The crash: a batch item with a null TagId makes the upstream Push throw (the documented unguarded
+        // path), so supervision restarts the alarm host. NO manual re-apply follows.
+        alarmHost.Tell(new VirtualTagActor.DependencyValueBatch(new[]
+        {
+            new VirtualTagActor.DependencyValueChanged(null!, 1, DateTime.UtcNow),
+        }));
+        mux.FishForMessage<DependencyMuxActor.UnregisterInterest>(u => u.Subscriber.Equals(alarmHost), Timeout);
+
+        // The parent re-applies on the restart signal: interest is registered again by the new incarnation.
+        mux.FishForMessage<DependencyMuxActor.RegisterInterest>(r => r.TagRefs.Contains("M.T"), Timeout);
+
+        // Inputs work: a dependency push activates the alarm.
+        alarmHost.Tell(new VirtualTagActor.DependencyValueChanged("M.T", 99, DateTime.UtcNow));
+        publish.FishForMessage<OpcUaPublishActor.AlarmStateUpdate>(
+            m => m.AlarmNodeId == "alm-1" && m.State.Active && !m.State.Acknowledged, Timeout);
+
+        // Acks work: the restarted engine owns alm-1 again, so the command is not filtered out as unowned.
+        alarmHost.Tell(new AlarmCommand(AlarmId: "alm-1", Operation: "Acknowledge", User: "alice", Comment: null, UnshelveAtUtc: null));
+        publish.FishForMessage<OpcUaPublishActor.AlarmStateUpdate>(
+            m => m.AlarmNodeId == "alm-1" && m.State.Acknowledged, Timeout)
+            .State.Acknowledged.ShouldBeTrue();
+    }
+}

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/Drivers/DriverHostActorSourceRemovedTests.cs

@@ -0,0 +1,224 @@
+using Akka.Actor;
+using Akka.TestKit;
+using Microsoft.EntityFrameworkCore;
+using Shouldly;
+using Xunit;
+using ZB.MOM.WW.OtOpcUa.Commons.Engines;
+using ZB.MOM.WW.OtOpcUa.Commons.Messages.Admin;
+using ZB.MOM.WW.OtOpcUa.Commons.OpcUa;
+using ZB.MOM.WW.OtOpcUa.Commons.Types;
+using ZB.MOM.WW.OtOpcUa.Configuration;
+using ZB.MOM.WW.OtOpcUa.Runtime.Drivers;
+using ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
+using ZB.MOM.WW.OtOpcUa.Runtime.VirtualTags;
+
+namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Drivers;
+
+/// <summary>
+///     The dependency mux retains a ref's last value for replay-opting subscribers and evicts a driver's values on
+///     <see cref="DependencyMuxActor.SourceRemoved"/>. Before this, nothing in production sent that message, so a
+///     driver removed by a deploy kept its last Good in the mux and replayed it to the next replay-opting
+///     subscriber as if it were current. The host now sends it wherever a driver OBJECT ends (deploy removal,
+///     config-change respawn, operator Restart), never on a supervision restart, and drops value bursts from the
+///     child incarnations it has already stopped so a late burst cannot re-retain what was evicted. All through a
+///     REAL mux and real driver children.
+/// </summary>
+public sealed class DriverHostActorSourceRemovedTests : RuntimeActorTestBase
+{
+    private static readonly NodeId TestNode = NodeId.Parse("driver-sourceremoved-test");
+    private static readonly RevisionHash RevA = RevisionHash.Parse(new string('a', 64));
+    private static readonly RevisionHash RevB = RevisionHash.Parse(new string('b', 64));
+    private const string Path1 = "Plant/Modbus/dev1/speed";
+    private const string Path2 = "Plant/Modbus2/dev2/flow";
+    private static readonly (string, string, string, string, string?)[] NoRefs = [];
+
+    private static readonly (string Driver, string DriverName, string Device, string Tag, string DataType, bool Writable)[] BothDrivers =
+    [
+        ("DI-1", "Modbus", "dev1", "speed", "Double", false),
+        ("DI-2", "Modbus2", "dev2", "flow", "Double", false),
+    ];
+
+    private IActorRef _host = ActorRefs.Nobody;
+    private IActorRef _mux = ActorRefs.Nobody;
+    private TestProbe _coordinator = null!;
+    private TestProbe _holder = null!;
+    private RecordingSubscribableDriverFactory _factory = null!;
+    private IDbContextFactory<OtOpcUaConfigDbContext> _db = null!;
+
+    /// <summary>Spawns a host over a real mux with two enabled drivers, waits for both children to subscribe,
+    /// registers a replay-opting holder (so the mux retains both refs) and publishes one value per driver.</summary>
+    private void ArrangeTwoDriversWithRetainedValues()
+    {
+        _db = NewInMemoryDbFactory();
+        var revA = V3ArtifactSeeder.SeedV3Deployment(_db, RevA, BothDrivers, NoRefs, enabled: true);
+        _factory = new RecordingSubscribableDriverFactory("Modbus");
+        _mux = Sys.ActorOf(DependencyMuxActor.Props());
+        _coordinator = CreateTestProbe();
+        _holder = CreateTestProbe();
+        _host = Sys.ActorOf(DriverHostActor.Props(
+            _db, TestNode, _coordinator.Ref,
+            driverFactory: _factory,
+            localRoles: new HashSet<string> { "driver" },
+            dependencyMux: _mux,
+            virtualTagEvaluator: NullVirtualTagEvaluator.Instance,
+            virtualTagHostOverride: CreateTestProbe().Ref));
+        V3ArtifactSeeder.Dispatch(_host, _coordinator, revA, RevA);
+        AwaitAssert(() => _factory.Created.Count.ShouldBe(2), PresenceBudget);
+        // The spawn hands a child the PREVIOUS apply's set (empty on the first deploy) and the bulk push then
+        // re-subscribes with the real one; a value fired before that lands is lost, so wait for the real set.
+        AwaitSubscribed("DI-1", Path1);
+        AwaitSubscribed("DI-2", Path2);
+
+        _mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { Path1, Path2 }, _holder.Ref, ReplayRetained: true));
+        Driver("DI-1").FireDataChange(Path1, 11.0);
+        Driver("DI-2").FireDataChange(Path2, 22.0);
+        // One fish for both: the two drivers' bursts arrive in either order, and fishing for one at a time would
+        // discard the other.
+        var pending = new HashSet<string>(StringComparer.Ordinal) { Path1, Path2 };
+        _holder.FishForMessage<VirtualTagActor.DependencyValueBatch>(b =>
+        {
+            foreach (var i in b.Items)
+                if ((i.TagId == Path1 && Equals(i.Value, 11.0)) || (i.TagId == Path2 && Equals(i.Value, 22.0)))
+                    pending.Remove(i.TagId);
+            return pending.Count == 0;
+        }, PresenceBudget);
+    }
+
+    /// <summary>Waits until the holder has been delivered <paramref name="value"/> for <paramref name="path"/>.</summary>
+    private void AwaitHolder(string path, object value) =>
+        _holder.FishForMessage<VirtualTagActor.DependencyValueBatch>(
+            b => b.Items.Any(i => i.TagId == path && Equals(i.Value, value)), PresenceBudget);
+
+    private void AwaitSubscribed(string driverId, string path) =>
+        AwaitAssert(() => (Driver(driverId).LastSubscribedRefs ?? []).ShouldContain(path), PresenceBudget);
+
+    private RecordingSubscribableDriver Driver(string id) => _factory.Created.Last(d => d.DriverInstanceId == id);
+
+    /// <summary>A fresh replay-opting subscriber for both refs; returns what it was replayed, keyed by ref.</summary>
+    private Dictionary<string, VirtualTagActor.DependencyValueChanged> Replayed()
+    {
+        var probe = CreateTestProbe();
+        _mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { Path1, Path2 }, probe.Ref, ReplayRetained: true));
+        return probe.ExpectMsg<VirtualTagActor.DependencyValueBatch>(PresenceBudget)
+            .Items.ToDictionary(i => i.TagId, StringComparer.Ordinal);
+    }
+
+    /// <summary>The live child actor for <paramref name="driverId"/> (named <c>drv-{id}-g{generation}</c>).</summary>
+    private IActorRef ChildOf(string driverId)
+    {
+        var probe = CreateTestProbe();
+        Sys.ActorSelection(_host.Path.ToStringWithoutAddress() + "/drv-" + driverId + "-g*").Tell(new Identify(driverId), probe.Ref);
+        return probe.ExpectMsg<ActorIdentity>(PresenceBudget).Subject.ShouldNotBeNull();
+    }
+
+    private static void ShouldNotBeTheOldGood(Dictionary<string, VirtualTagActor.DependencyValueChanged> replay, string path, object oldValue)
+    {
+        if (!replay.TryGetValue(path, out var item)) return;
+        (Equals(item.Value, oldValue) && OpcUaStatus.IsGood(item.StatusCode))
+            .ShouldBeFalse($"the stopped incarnation's last Good for {path} was replayed as current");
+    }
+
+    [Fact]
+    public void A_driver_removed_by_a_deploy_is_no_longer_replayed_while_a_still_present_driver_is()
+    {
+        ArrangeTwoDriversWithRetainedValues();
+
+        // Positive control before the removal: both values replay.
+        Replayed().Keys.ShouldBe(new[] { Path1, Path2 }, ignoreOrder: true);
+
+        // Deploy B drops DI-1 (StopChild).
+        var revB = V3ArtifactSeeder.SeedV3Deployment(_db, RevB, [("DI-2", "Modbus2", "dev2", "flow", "Double", false)], NoRefs, enabled: true);
+        V3ArtifactSeeder.Dispatch(_host, _coordinator, revB, RevB);
+
+        AwaitAssert(
+            () => Replayed().Keys.ToArray().ShouldBe(new[] { Path2 }, "DI-1 was removed: its last Good must not replay; DI-2's must"),
+            PresenceBudget);
+    }
+
+    /// <summary>Review I-1: a burst the removed child sent after the eviction (it stops asynchronously, so it keeps
+    /// publishing through the rest of the apply) must be dropped by the host, not forwarded to re-retain the value.</summary>
+    [Fact]
+    public void A_late_burst_from_a_removed_child_does_not_re_retain_its_value()
+    {
+        ArrangeTwoDriversWithRetainedValues();
+        var oldChild = ChildOf("DI-1");
+        // Hold the removed child in its PostStop so it has NOT terminated when the stale burst lands; a terminated
+        // actor cannot send, and the host forgets a retired child on Terminated.
+        var gate = new TaskCompletionSource(TaskCreationOptions.RunContinuationsAsynchronously);
+        Driver("DI-1").ShutdownGate = gate;
+
+        var revB = V3ArtifactSeeder.SeedV3Deployment(_db, RevB, [("DI-2", "Modbus2", "dev2", "flow", "Double", false)], NoRefs, enabled: true);
+        V3ArtifactSeeder.Dispatch(_host, _coordinator, revB, RevB);
+
+        // The stale burst, sent AS the stopped child.
+        _host.Tell(ValuePublishes.Single("DI-1", Path1, 11.0, OpcUaStatus.Good, DateTime.UtcNow), oldChild);
+        // Positive control / ordering fence: a live driver's value sent after it is delivered, so the stale burst
+        // has been processed by the host by the time we look.
+        Driver("DI-2").FireDataChange(Path2, 23.0);
+        AwaitHolder(Path2, 23.0);
+
+        Replayed().Keys.ToArray().ShouldBe(new[] { Path2 }, "the removed child's late burst re-retained its last Good");
+        gate.TrySetResult();
+    }
+
+    /// <summary>Review m-3: a config edit respawns the driver (stop + spawn); the OLD object's value is evicted, and
+    /// the NEW driver's value replays (positive control).</summary>
+    [Fact]
+    public void A_config_change_respawn_evicts_the_old_value_and_the_new_drivers_value_replays()
+    {
+        ArrangeTwoDriversWithRetainedValues();
+
+        // Adding a tag to DI-1 changes its merged DriverConfig, which respawns it.
+        var revB = V3ArtifactSeeder.SeedV3Deployment(_db, RevB,
+            [.. BothDrivers, ("DI-1", "Modbus", "dev1", "temp", "Double", false)], NoRefs, enabled: true);
+        V3ArtifactSeeder.Dispatch(_host, _coordinator, revB, RevB);
+        AwaitAssert(() => _factory.Created.Count.ShouldBe(3), PresenceBudget);
+        var respawned = Driver("DI-1");
+        // The spawn hands the child the PREVIOUS apply's set, then the bulk push re-subscribes with the new one
+        // (which names the added tag); a value fired between the two can be lost, so wait for the final set.
+        AwaitAssert(() => (respawned.LastSubscribedRefs ?? []).ShouldContain("Plant/Modbus/dev1/temp"), PresenceBudget);
+
+        ShouldNotBeTheOldGood(Replayed(), Path1, 11.0);
+
+        respawned.FireDataChange(Path1, 33.0);
+        AwaitHolder(Path1, 33.0);
+        var replay = Replayed();
+        replay[Path1].Value.ShouldBe(33.0, "the respawned driver's own value replays");
+        replay[Path2].Value.ShouldBe(22.0, "the untouched driver is unaffected");
+    }
+
+    /// <summary>Review m-3: an operator Restart builds a new driver object; the old one's value is evicted and the
+    /// new one's replays.</summary>
+    [Fact]
+    public void An_operator_restart_evicts_the_old_value_and_the_new_drivers_value_replays()
+    {
+        ArrangeTwoDriversWithRetainedValues();
+
+        _host.Tell(new RestartDriver("c1", "DI-1", "tester", Guid.NewGuid()));
+        AwaitAssert(() => _factory.Created.Count.ShouldBe(3), PresenceBudget);
+        var restarted = Driver("DI-1");
+        AwaitSubscribed("DI-1", Path1);
+
+        ShouldNotBeTheOldGood(Replayed(), Path1, 11.0);
+
+        restarted.FireDataChange(Path1, 44.0);
+        AwaitHolder(Path1, 44.0);
+        Replayed()[Path1].Value.ShouldBe(44.0);
+    }
+
+    [Fact]
+    public void A_supervision_restart_does_not_evict_the_drivers_retained_value()
+    {
+        ArrangeTwoDriversWithRetainedValues();
+        var di1 = Driver("DI-1");
+
+        // Crash DI-1's child: Akka restarts it around the SAME driver object, and the host reprovisions it.
+        di1.FireMalformedDataChange(Path1);
+        AwaitAssert(() => di1.InitializeCount.ShouldBe(2), PresenceBudget);
+        _factory.Created.Count.ShouldBe(2, "a supervision restart re-uses the driver; no new one is created");
+
+        var replay = Replayed();
+        replay.Keys.ShouldBe(new[] { Path1, Path2 }, ignoreOrder: true,
+            "a restart is the same driver instance: its last value stays replayable");
+    }
+}

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/Drivers/DriverInstanceActorBatchForwardTests.cs

@@ -110,7 +110,7 @@ public sealed class DriverInstanceActorBatchForwardTests : RuntimeActorTestBase
         batch.Items.Count.ShouldBe(2);
 
         // The per-tag raises for the SAME burst produced nothing extra — no second batch, no
-        // AttributeValuePublished. Anything here means both handlers were attached.
+        // single-item batch. Anything here means both handlers were attached.
         parent.ExpectNoMsg(TimeSpan.FromMilliseconds(400));
     }
 

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/Drivers/StubDrivers.cs

@@ -279,11 +279,18 @@ internal sealed class RecordingSubscribableDriver : IDriver, ISubscribable, IAla
     /// <inheritdoc />
     public Task ReinitializeAsync(string driverConfigJson, CancellationToken cancellationToken) => Task.CompletedTask;
 
+    /// <summary>
+    ///     When set, <see cref="ShutdownAsync"/> awaits it (the actor bounds the wait), which keeps a stopping child
+    ///     ALIVE in its PostStop for as long as the test wants — the window in which a stopped child can still
+    ///     have messages reaching its parent.
+    /// </summary>
+    public TaskCompletionSource? ShutdownGate { get; set; }
+
     /// <inheritdoc />
     public Task ShutdownAsync(CancellationToken cancellationToken)
     {
         Interlocked.Increment(ref _shutdownCount);
-        return Task.CompletedTask;
+        return ShutdownGate?.Task ?? Task.CompletedTask;
     }
 
     /// <inheritdoc />

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/Harness/ValuePublishes.cs

@@ -5,8 +5,8 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Harness;
 /// <summary>
 ///     Builds the ONE message shape a driver child sends for its values — a
 ///     <see cref="DriverInstanceActor.AttributeValueBatchPublished"/> — for tests that push a single value into the
-///     dependency mux by hand. The per-value <see cref="DriverInstanceActor.AttributeValuePublished"/> has no
-///     production sender (02/U4); tests should not keep one alive either.
+///     dependency mux by hand. There is no per-value message: the former per-value record had no production
+///     sender and was deleted (02/U4).
 /// </summary>
 public static class ValuePublishes
 {

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/Historian/ContinuousHistorizationRecorderTests.cs

@@ -19,7 +19,7 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.Historian;
 /// <remarks>
 ///     Adapted from the plan's Task 17: the recorder handles the REAL fan-out message the mux emits —
 ///     <see cref="VirtualTagActor.DependencyValueChanged"/> (TagId/Value/TimestampUtc/StatusCode) —
-///     not <c>DriverInstanceActor.AttributeValuePublished</c>. The typed-value / real-status behaviour is
+///     not the driver-side <c>DriverInstanceActor.AttributeValueBatchPublished</c>. The typed-value / real-status behaviour is
 ///     pinned in <see cref="ContinuousHistorizationRecorderTypedValueTests"/>.
 /// </remarks>
 public sealed class ContinuousHistorizationRecorderTests : TestKit

tests/Server/ZB.MOM.WW.OtOpcUa.Runtime.Tests/VirtualTags/DependencyMuxActorTests.cs

@@ -10,9 +10,9 @@ namespace ZB.MOM.WW.OtOpcUa.Runtime.Tests.VirtualTags;
 
 public sealed class DependencyMuxActorTests : RuntimeActorTestBase
 {
-    /// <summary>Verifies that AttributeValuePublished messages are forwarded only to subscribers interested in the specific reference.</summary>
+    /// <summary>Verifies that a published value is forwarded only to subscribers interested in the specific reference.</summary>
     [Fact]
-    public void AttributeValuePublished_is_forwarded_only_to_subscribers_interested_in_that_ref()
+    public void A_published_value_is_forwarded_only_to_subscribers_interested_in_that_ref()
     {
         var mux = Sys.ActorOf(DependencyMuxActor.Props());
         var subA = CreateTestProbe();
@@ -21,9 +21,9 @@ public sealed class DependencyMuxActorTests : RuntimeActorTestBase
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-1", "tag-2" }, subA.Ref));
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-2", "tag-3" }, subB.Ref));
 
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-1", 10, OpcUaStatus.Good, DateTime.UtcNow));
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-3", 30, OpcUaStatus.Good, DateTime.UtcNow));
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-2", 20, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-1", 10, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-3", 30, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-2", 20, OpcUaStatus.Good, DateTime.UtcNow));
 
         // The mux emits ONLY batches — the per-value comms-loss ingress fans out as a single-item batch.
         // subA hears tag-1 + tag-2.
@@ -54,7 +54,7 @@ public sealed class DependencyMuxActorTests : RuntimeActorTestBase
         var sub = CreateTestProbe();
 
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-1" }, sub.Ref));
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "nobody-cares", 99, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "nobody-cares", 99, OpcUaStatus.Good, DateTime.UtcNow));
 
         sub.ExpectNoMsg(TimeSpan.FromMilliseconds(200));
     }
@@ -67,11 +67,11 @@ public sealed class DependencyMuxActorTests : RuntimeActorTestBase
         var sub = CreateTestProbe();
 
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-1" }, sub.Ref));
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-1", 10, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-1", 10, OpcUaStatus.Good, DateTime.UtcNow));
         sub.ExpectMsg<VirtualTagActor.DependencyValueBatch>();
 
         mux.Tell(new DependencyMuxActor.UnregisterInterest(sub.Ref));
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-1", 20, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-1", 20, OpcUaStatus.Good, DateTime.UtcNow));
         sub.ExpectNoMsg(TimeSpan.FromMilliseconds(200));
     }
 
@@ -85,10 +85,10 @@ public sealed class DependencyMuxActorTests : RuntimeActorTestBase
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-1" }, sub.Ref));
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-2" }, sub.Ref)); // replaces tag-1
 
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-1", 10, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-1", 10, OpcUaStatus.Good, DateTime.UtcNow));
         sub.ExpectNoMsg(TimeSpan.FromMilliseconds(200));
 
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "tag-2", 20, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "tag-2", 20, OpcUaStatus.Good, DateTime.UtcNow));
         sub.ExpectMsg<VirtualTagActor.DependencyValueBatch>().Items.ShouldHaveSingleItem().TagId.ShouldBe("tag-2");
     }
 
@@ -110,18 +110,18 @@ public sealed class DependencyMuxActorTests : RuntimeActorTestBase
         // gets dropped (no subscriber for "ref-a" yet), so we re-publish each pass.
         AwaitAssert(() =>
         {
-            mux.Tell(new DriverInstanceActor.AttributeValuePublished(
+            mux.Tell(ValuePublishes.Single(
                 "driver-1", "ref-a", 10, OpcUaStatus.Good, DateTime.UtcNow));
             parent.ExpectMsg<VirtualTagActor.EvaluationResult>(TimeSpan.FromMilliseconds(200))
                 .Value.ShouldBe(10);
         }, duration: PresenceBudget);
 
         // From here the actor is wired — second publish drives the sum.
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "ref-b", 32, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "ref-b", 32, OpcUaStatus.Good, DateTime.UtcNow));
         parent.ExpectMsg<VirtualTagActor.EvaluationResult>(PresenceBudget).Value.ShouldBe(42);
 
         // Unrelated ref shouldn't fire eval.
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished("driver-1", "ref-unrelated", 99, OpcUaStatus.Good, DateTime.UtcNow));
+        mux.Tell(ValuePublishes.Single("driver-1", "ref-unrelated", 99, OpcUaStatus.Good, DateTime.UtcNow));
         parent.ExpectNoMsg(TimeSpan.FromMilliseconds(200));
     }
 
@@ -221,7 +221,7 @@ public sealed class DependencyMuxActorTests : RuntimeActorTestBase
         var sub = CreateTestProbe();
         mux.Tell(new DependencyMuxActor.RegisterInterest(new[] { "tag-1" }, sub.Ref));
 
-        mux.Tell(new DriverInstanceActor.AttributeValuePublished(
+        mux.Tell(ValuePublishes.Single(
             "driver-1", "tag-1", 10, OpcUaStatus.Bad, DateTime.UtcNow));
 
         sub.ExpectMsg<VirtualTagActor.DependencyValueBatch>()